Commit | Line | Data |
---|---|---|
2e1df07d JA |
1 | /* |
2 | * fio - the flexible io tester | |
3 | * | |
4 | * Copyright (C) 2005 Jens Axboe <axboe@suse.de> | |
5 | * Copyright (C) 2006-2012 Jens Axboe <axboe@kernel.dk> | |
6 | * | |
7 | * The license below covers all files distributed with fio unless otherwise | |
8 | * noted in the file itself. | |
9 | * | |
10 | * This program is free software; you can redistribute it and/or modify | |
11 | * it under the terms of the GNU General Public License version 2 as | |
12 | * published by the Free Software Foundation. | |
13 | * | |
14 | * This program is distributed in the hope that it will be useful, | |
15 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
16 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the | |
17 | * GNU General Public License for more details. | |
18 | * | |
19 | * You should have received a copy of the GNU General Public License | |
20 | * along with this program; if not, write to the Free Software | |
fa07eaa6 | 21 | * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301, USA. |
2e1df07d JA |
22 | * |
23 | */ | |
24 | #include <unistd.h> | |
2e1df07d | 25 | #include <string.h> |
2e1df07d | 26 | #include <signal.h> |
2e1df07d | 27 | #include <assert.h> |
e43606c2 | 28 | #include <inttypes.h> |
2e1df07d JA |
29 | #include <sys/stat.h> |
30 | #include <sys/wait.h> | |
ff6bb260 | 31 | #include <math.h> |
c06379a6 | 32 | #include <pthread.h> |
2e1df07d JA |
33 | |
34 | #include "fio.h" | |
2e1df07d JA |
35 | #include "smalloc.h" |
36 | #include "verify.h" | |
2e1df07d JA |
37 | #include "diskutil.h" |
38 | #include "cgroup.h" | |
39 | #include "profile.h" | |
40 | #include "lib/rand.h" | |
f7690c4a | 41 | #include "lib/memalign.h" |
2e1df07d | 42 | #include "server.h" |
44404c5a | 43 | #include "lib/getrusage.h" |
f2a2ce0e | 44 | #include "idletime.h" |
002fe734 | 45 | #include "err.h" |
a9da8ab2 | 46 | #include "workqueue.h" |
e81ecca3 | 47 | #include "lib/mountcheck.h" |
40511301 | 48 | #include "rate-submit.h" |
a39fb9ea | 49 | #include "helper_thread.h" |
ae626d4e | 50 | #include "pshared.h" |
07776127 | 51 | #include "zone-dist.h" |
2e1df07d | 52 | |
971caeb1 | 53 | static struct fio_sem *startup_sem; |
2e1df07d | 54 | static struct flist_head *cgroup_list; |
c3dc516a | 55 | static struct cgroup_mnt *cgroup_mnt; |
2e1df07d | 56 | static int exit_value; |
f1867a7f | 57 | static volatile bool fio_abort; |
3a5f6bde JA |
58 | static unsigned int nr_process = 0; |
59 | static unsigned int nr_thread = 0; | |
2e1df07d | 60 | |
6eaf09d6 | 61 | struct io_log *agg_io_log[DDIR_RWDIR_CNT]; |
2e1df07d | 62 | |
a3efc919 JA |
63 | int groupid = 0; |
64 | unsigned int thread_number = 0; | |
2fbe1e19 JA |
65 | unsigned int nr_segments = 0; |
66 | unsigned int cur_segment = 0; | |
108fea77 | 67 | unsigned int stat_number = 0; |
a3efc919 JA |
68 | int temp_stall_ts; |
69 | unsigned long done_secs = 0; | |
a38c70a8 BVA |
70 | #ifdef PTHREAD_ERRORCHECK_MUTEX_INITIALIZER_NP |
71 | pthread_mutex_t overlap_check = PTHREAD_ERRORCHECK_MUTEX_INITIALIZER_NP; | |
72 | #else | |
c06379a6 | 73 | pthread_mutex_t overlap_check = PTHREAD_MUTEX_INITIALIZER; |
a38c70a8 | 74 | #endif |
a47591e4 | 75 | |
2e1df07d JA |
76 | #define JOB_START_TIMEOUT (5 * 1000) |
77 | ||
78 | static void sig_int(int sig) | |
79 | { | |
2fbe1e19 | 80 | if (nr_segments) { |
2e1df07d JA |
81 | if (is_backend) |
82 | fio_server_got_signal(sig); | |
83 | else { | |
84 | log_info("\nfio: terminating on signal %d\n", sig); | |
e411c301 | 85 | log_info_flush(); |
2e1df07d JA |
86 | exit_value = 128; |
87 | } | |
88 | ||
64402a8a | 89 | fio_terminate_threads(TERMINATE_ALL, TERMINATE_ALL); |
2e1df07d JA |
90 | } |
91 | } | |
92 | ||
d2235e56 | 93 | void sig_show_status(int sig) |
4c6d91e8 JA |
94 | { |
95 | show_running_run_stats(); | |
96 | } | |
97 | ||
2e1df07d JA |
98 | static void set_sig_handlers(void) |
99 | { | |
100 | struct sigaction act; | |
101 | ||
102 | memset(&act, 0, sizeof(act)); | |
103 | act.sa_handler = sig_int; | |
104 | act.sa_flags = SA_RESTART; | |
105 | sigaction(SIGINT, &act, NULL); | |
106 | ||
107 | memset(&act, 0, sizeof(act)); | |
108 | act.sa_handler = sig_int; | |
109 | act.sa_flags = SA_RESTART; | |
110 | sigaction(SIGTERM, &act, NULL); | |
111 | ||
2f694507 BC |
112 | /* Windows uses SIGBREAK as a quit signal from other applications */ |
113 | #ifdef WIN32 | |
114 | memset(&act, 0, sizeof(act)); | |
115 | act.sa_handler = sig_int; | |
116 | act.sa_flags = SA_RESTART; | |
117 | sigaction(SIGBREAK, &act, NULL); | |
118 | #endif | |
119 | ||
4c6d91e8 JA |
120 | memset(&act, 0, sizeof(act)); |
121 | act.sa_handler = sig_show_status; | |
122 | act.sa_flags = SA_RESTART; | |
123 | sigaction(SIGUSR1, &act, NULL); | |
124 | ||
2e1df07d JA |
125 | if (is_backend) { |
126 | memset(&act, 0, sizeof(act)); | |
127 | act.sa_handler = sig_int; | |
128 | act.sa_flags = SA_RESTART; | |
129 | sigaction(SIGPIPE, &act, NULL); | |
130 | } | |
131 | } | |
132 | ||
133 | /* | |
134 | * Check if we are above the minimum rate given. | |
135 | */ | |
8b6a404c | 136 | static bool __check_min_rate(struct thread_data *td, struct timespec *now, |
e39c0676 | 137 | enum fio_ddir ddir) |
2e1df07d JA |
138 | { |
139 | unsigned long long bytes = 0; | |
140 | unsigned long iops = 0; | |
141 | unsigned long spent; | |
140a6888 SB |
142 | unsigned long long rate; |
143 | unsigned long long ratemin = 0; | |
2e1df07d JA |
144 | unsigned int rate_iops = 0; |
145 | unsigned int rate_iops_min = 0; | |
146 | ||
147 | assert(ddir_rw(ddir)); | |
148 | ||
149 | if (!td->o.ratemin[ddir] && !td->o.rate_iops_min[ddir]) | |
e39c0676 | 150 | return false; |
2e1df07d JA |
151 | |
152 | /* | |
153 | * allow a 2 second settle period in the beginning | |
154 | */ | |
155 | if (mtime_since(&td->start, now) < 2000) | |
e39c0676 | 156 | return false; |
2e1df07d JA |
157 | |
158 | iops += td->this_io_blocks[ddir]; | |
159 | bytes += td->this_io_bytes[ddir]; | |
160 | ratemin += td->o.ratemin[ddir]; | |
161 | rate_iops += td->o.rate_iops[ddir]; | |
162 | rate_iops_min += td->o.rate_iops_min[ddir]; | |
163 | ||
164 | /* | |
165 | * if rate blocks is set, sample is running | |
166 | */ | |
167 | if (td->rate_bytes[ddir] || td->rate_blocks[ddir]) { | |
168 | spent = mtime_since(&td->lastrate[ddir], now); | |
169 | if (spent < td->o.ratecycle) | |
e39c0676 | 170 | return false; |
2e1df07d | 171 | |
20e37e4a | 172 | if (td->o.rate[ddir] || td->o.ratemin[ddir]) { |
2e1df07d JA |
173 | /* |
174 | * check bandwidth specified rate | |
175 | */ | |
176 | if (bytes < td->rate_bytes[ddir]) { | |
140a6888 | 177 | log_err("%s: rate_min=%lluB/s not met, only transferred %lluB\n", |
420b104a | 178 | td->o.name, ratemin, bytes); |
e39c0676 | 179 | return true; |
2e1df07d | 180 | } else { |
49cba9b3 JA |
181 | if (spent) |
182 | rate = ((bytes - td->rate_bytes[ddir]) * 1000) / spent; | |
183 | else | |
184 | rate = 0; | |
185 | ||
2e1df07d JA |
186 | if (rate < ratemin || |
187 | bytes < td->rate_bytes[ddir]) { | |
140a6888 | 188 | log_err("%s: rate_min=%lluB/s not met, got %lluB/s\n", |
420b104a | 189 | td->o.name, ratemin, rate); |
e39c0676 | 190 | return true; |
2e1df07d JA |
191 | } |
192 | } | |
193 | } else { | |
194 | /* | |
195 | * checks iops specified rate | |
196 | */ | |
197 | if (iops < rate_iops) { | |
420b104a RE |
198 | log_err("%s: rate_iops_min=%u not met, only performed %lu IOs\n", |
199 | td->o.name, rate_iops, iops); | |
e39c0676 | 200 | return true; |
2e1df07d | 201 | } else { |
4c707a3b JA |
202 | if (spent) |
203 | rate = ((iops - td->rate_blocks[ddir]) * 1000) / spent; | |
204 | else | |
205 | rate = 0; | |
206 | ||
2e1df07d JA |
207 | if (rate < rate_iops_min || |
208 | iops < td->rate_blocks[ddir]) { | |
140a6888 | 209 | log_err("%s: rate_iops_min=%u not met, got %llu IOPS\n", |
420b104a | 210 | td->o.name, rate_iops_min, rate); |
e39c0676 | 211 | return true; |
2e1df07d JA |
212 | } |
213 | } | |
214 | } | |
215 | } | |
216 | ||
217 | td->rate_bytes[ddir] = bytes; | |
218 | td->rate_blocks[ddir] = iops; | |
219 | memcpy(&td->lastrate[ddir], now, sizeof(*now)); | |
e39c0676 | 220 | return false; |
2e1df07d JA |
221 | } |
222 | ||
8b6a404c | 223 | static bool check_min_rate(struct thread_data *td, struct timespec *now) |
2e1df07d | 224 | { |
e39c0676 | 225 | bool ret = false; |
2e1df07d | 226 | |
dde39f8c AD |
227 | for_each_rw_ddir(ddir) { |
228 | if (td->bytes_done[ddir]) | |
229 | ret |= __check_min_rate(td, now, ddir); | |
230 | } | |
2e1df07d JA |
231 | |
232 | return ret; | |
233 | } | |
234 | ||
235 | /* | |
236 | * When job exits, we can cancel the in-flight IO if we are using async | |
237 | * io. Attempt to do so. | |
238 | */ | |
239 | static void cleanup_pending_aio(struct thread_data *td) | |
240 | { | |
2e1df07d JA |
241 | int r; |
242 | ||
243 | /* | |
244 | * get immediately available events, if any | |
245 | */ | |
55312f9f | 246 | r = io_u_queued_complete(td, 0); |
2e1df07d JA |
247 | |
248 | /* | |
249 | * now cancel remaining active events | |
250 | */ | |
251 | if (td->io_ops->cancel) { | |
2ae0b204 JA |
252 | struct io_u *io_u; |
253 | int i; | |
2e1df07d | 254 | |
2ae0b204 JA |
255 | io_u_qiter(&td->io_u_all, io_u, i) { |
256 | if (io_u->flags & IO_U_F_FLIGHT) { | |
2e1df07d JA |
257 | r = td->io_ops->cancel(td, io_u); |
258 | if (!r) | |
259 | put_io_u(td, io_u); | |
260 | } | |
261 | } | |
262 | } | |
263 | ||
264 | if (td->cur_depth) | |
55312f9f | 265 | r = io_u_queued_complete(td, td->cur_depth); |
2e1df07d JA |
266 | } |
267 | ||
268 | /* | |
269 | * Helper to handle the final sync of a file. Works just like the normal | |
270 | * io path, just does everything sync. | |
271 | */ | |
e39c0676 | 272 | static bool fio_io_sync(struct thread_data *td, struct fio_file *f) |
2e1df07d JA |
273 | { |
274 | struct io_u *io_u = __get_io_u(td); | |
d3b07186 | 275 | enum fio_q_status ret; |
2e1df07d JA |
276 | |
277 | if (!io_u) | |
e39c0676 | 278 | return true; |
2e1df07d JA |
279 | |
280 | io_u->ddir = DDIR_SYNC; | |
281 | io_u->file = f; | |
00615bfb | 282 | io_u_set(td, io_u, IO_U_F_NO_FILE_PUT); |
2e1df07d JA |
283 | |
284 | if (td_io_prep(td, io_u)) { | |
285 | put_io_u(td, io_u); | |
e39c0676 | 286 | return true; |
2e1df07d JA |
287 | } |
288 | ||
289 | requeue: | |
290 | ret = td_io_queue(td, io_u); | |
18f6165e BVA |
291 | switch (ret) { |
292 | case FIO_Q_QUEUED: | |
a80cb54b | 293 | td_io_commit(td); |
55312f9f | 294 | if (io_u_queued_complete(td, 1) < 0) |
e39c0676 | 295 | return true; |
18f6165e BVA |
296 | break; |
297 | case FIO_Q_COMPLETED: | |
2e1df07d JA |
298 | if (io_u->error) { |
299 | td_verror(td, io_u->error, "td_io_queue"); | |
e39c0676 | 300 | return true; |
2e1df07d JA |
301 | } |
302 | ||
55312f9f | 303 | if (io_u_sync_complete(td, io_u) < 0) |
e39c0676 | 304 | return true; |
18f6165e BVA |
305 | break; |
306 | case FIO_Q_BUSY: | |
a80cb54b | 307 | td_io_commit(td); |
2e1df07d JA |
308 | goto requeue; |
309 | } | |
310 | ||
e39c0676 | 311 | return false; |
2e1df07d | 312 | } |
a3efc919 | 313 | |
61ee0f86 JA |
314 | static int fio_file_fsync(struct thread_data *td, struct fio_file *f) |
315 | { | |
00615bfb | 316 | int ret, ret2; |
61ee0f86 JA |
317 | |
318 | if (fio_file_open(f)) | |
319 | return fio_io_sync(td, f); | |
320 | ||
321 | if (td_io_open_file(td, f)) | |
322 | return 1; | |
323 | ||
324 | ret = fio_io_sync(td, f); | |
00615bfb SW |
325 | ret2 = 0; |
326 | if (fio_file_open(f)) | |
327 | ret2 = td_io_close_file(td, f); | |
328 | return (ret || ret2); | |
61ee0f86 JA |
329 | } |
330 | ||
8b6a404c | 331 | static inline void __update_ts_cache(struct thread_data *td) |
2e1df07d | 332 | { |
8b6a404c | 333 | fio_gettime(&td->ts_cache, NULL); |
2e1df07d JA |
334 | } |
335 | ||
8b6a404c | 336 | static inline void update_ts_cache(struct thread_data *td) |
2e1df07d | 337 | { |
8b6a404c VF |
338 | if ((++td->ts_cache_nr & td->ts_cache_mask) == td->ts_cache_mask) |
339 | __update_ts_cache(td); | |
2e1df07d JA |
340 | } |
341 | ||
8b6a404c | 342 | static inline bool runtime_exceeded(struct thread_data *td, struct timespec *t) |
2e1df07d JA |
343 | { |
344 | if (in_ramp_time(td)) | |
e39c0676 | 345 | return false; |
2e1df07d | 346 | if (!td->o.timeout) |
e39c0676 | 347 | return false; |
0de5b26f | 348 | if (utime_since(&td->epoch, t) >= td->o.timeout) |
e39c0676 | 349 | return true; |
2e1df07d | 350 | |
e39c0676 | 351 | return false; |
2e1df07d JA |
352 | } |
353 | ||
95603b74 BF |
354 | /* |
355 | * We need to update the runtime consistently in ms, but keep a running | |
356 | * tally of the current elapsed time in microseconds for sub millisecond | |
357 | * updates. | |
358 | */ | |
359 | static inline void update_runtime(struct thread_data *td, | |
360 | unsigned long long *elapsed_us, | |
361 | const enum fio_ddir ddir) | |
362 | { | |
d5c8ea29 JA |
363 | if (ddir == DDIR_WRITE && td_write(td) && td->o.verify_only) |
364 | return; | |
365 | ||
95603b74 BF |
366 | td->ts.runtime[ddir] -= (elapsed_us[ddir] + 999) / 1000; |
367 | elapsed_us[ddir] += utime_since_now(&td->start); | |
368 | td->ts.runtime[ddir] += (elapsed_us[ddir] + 999) / 1000; | |
369 | } | |
370 | ||
e39c0676 JA |
371 | static bool break_on_this_error(struct thread_data *td, enum fio_ddir ddir, |
372 | int *retptr) | |
2e1df07d JA |
373 | { |
374 | int ret = *retptr; | |
375 | ||
376 | if (ret < 0 || td->error) { | |
8b28bd41 DM |
377 | int err = td->error; |
378 | enum error_type_bit eb; | |
2e1df07d JA |
379 | |
380 | if (ret < 0) | |
381 | err = -ret; | |
2e1df07d | 382 | |
8b28bd41 DM |
383 | eb = td_error_type(ddir, err); |
384 | if (!(td->o.continue_on_error & (1 << eb))) | |
e39c0676 | 385 | return true; |
2e1df07d | 386 | |
8b28bd41 | 387 | if (td_non_fatal_error(td, eb, err)) { |
2e1df07d JA |
388 | /* |
389 | * Continue with the I/Os in case of | |
390 | * a non fatal error. | |
391 | */ | |
392 | update_error_count(td, err); | |
393 | td_clear_error(td); | |
394 | *retptr = 0; | |
e39c0676 | 395 | return false; |
2e1df07d JA |
396 | } else if (td->o.fill_device && err == ENOSPC) { |
397 | /* | |
398 | * We expect to hit this error if | |
399 | * fill_device option is set. | |
400 | */ | |
401 | td_clear_error(td); | |
ebea2133 | 402 | fio_mark_td_terminate(td); |
e39c0676 | 403 | return true; |
2e1df07d JA |
404 | } else { |
405 | /* | |
406 | * Stop the I/O in case of a fatal | |
407 | * error. | |
408 | */ | |
409 | update_error_count(td, err); | |
e39c0676 | 410 | return true; |
2e1df07d JA |
411 | } |
412 | } | |
413 | ||
e39c0676 | 414 | return false; |
2e1df07d JA |
415 | } |
416 | ||
c97f1ad6 JA |
417 | static void check_update_rusage(struct thread_data *td) |
418 | { | |
419 | if (td->update_rusage) { | |
420 | td->update_rusage = 0; | |
421 | update_rusage_stat(td); | |
971caeb1 | 422 | fio_sem_up(td->rusage_sem); |
c97f1ad6 JA |
423 | } |
424 | } | |
425 | ||
8b6a404c | 426 | static int wait_for_completions(struct thread_data *td, struct timespec *time) |
69fea81e JA |
427 | { |
428 | const int full = queue_full(td); | |
429 | int min_evts = 0; | |
430 | int ret; | |
431 | ||
6be06c46 JA |
432 | if (td->flags & TD_F_REGROW_LOGS) |
433 | return io_u_quiesce(td); | |
1fed2080 | 434 | |
69fea81e JA |
435 | /* |
436 | * if the queue is full, we MUST reap at least 1 event | |
437 | */ | |
82407585 | 438 | min_evts = min(td->o.iodepth_batch_complete_min, td->cur_depth); |
4b157ac6 | 439 | if ((full && !min_evts) || !td->o.iodepth_batch_complete_min) |
69fea81e JA |
440 | min_evts = 1; |
441 | ||
90711edc | 442 | if (time && should_check_rate(td)) |
69fea81e JA |
443 | fio_gettime(time, NULL); |
444 | ||
445 | do { | |
55312f9f | 446 | ret = io_u_queued_complete(td, min_evts); |
69fea81e JA |
447 | if (ret < 0) |
448 | break; | |
449 | } while (full && (td->cur_depth > td->o.iodepth_low)); | |
450 | ||
451 | return ret; | |
452 | } | |
453 | ||
e9d512d8 JA |
454 | int io_queue_event(struct thread_data *td, struct io_u *io_u, int *ret, |
455 | enum fio_ddir ddir, uint64_t *bytes_issued, int from_verify, | |
8b6a404c | 456 | struct timespec *comp_time) |
e9d512d8 | 457 | { |
e9d512d8 JA |
458 | switch (*ret) { |
459 | case FIO_Q_COMPLETED: | |
460 | if (io_u->error) { | |
461 | *ret = -io_u->error; | |
462 | clear_io_u(td, io_u); | |
463 | } else if (io_u->resid) { | |
5fff9543 | 464 | long long bytes = io_u->xfer_buflen - io_u->resid; |
e9d512d8 JA |
465 | struct fio_file *f = io_u->file; |
466 | ||
467 | if (bytes_issued) | |
468 | *bytes_issued += bytes; | |
469 | ||
470 | if (!from_verify) | |
94a78d52 | 471 | trim_io_piece(io_u); |
e9d512d8 JA |
472 | |
473 | /* | |
474 | * zero read, fail | |
475 | */ | |
476 | if (!bytes) { | |
477 | if (!from_verify) | |
478 | unlog_io_piece(td, io_u); | |
479 | td_verror(td, EIO, "full resid"); | |
480 | put_io_u(td, io_u); | |
481 | break; | |
482 | } | |
483 | ||
484 | io_u->xfer_buflen = io_u->resid; | |
485 | io_u->xfer_buf += bytes; | |
486 | io_u->offset += bytes; | |
487 | ||
488 | if (ddir_rw(io_u->ddir)) | |
489 | td->ts.short_io_u[io_u->ddir]++; | |
490 | ||
e9d512d8 JA |
491 | if (io_u->offset == f->real_file_size) |
492 | goto sync_done; | |
493 | ||
494 | requeue_io_u(td, &io_u); | |
495 | } else { | |
496 | sync_done: | |
90711edc | 497 | if (comp_time && should_check_rate(td)) |
e9d512d8 JA |
498 | fio_gettime(comp_time, NULL); |
499 | ||
500 | *ret = io_u_sync_complete(td, io_u); | |
501 | if (*ret < 0) | |
502 | break; | |
503 | } | |
a0e3e5a6 | 504 | |
cf2c8d52 JA |
505 | if (td->flags & TD_F_REGROW_LOGS) |
506 | regrow_logs(td); | |
507 | ||
a0e3e5a6 EW |
508 | /* |
509 | * when doing I/O (not when verifying), | |
510 | * check for any errors that are to be ignored | |
511 | */ | |
512 | if (!from_verify) | |
513 | break; | |
514 | ||
e9d512d8 JA |
515 | return 0; |
516 | case FIO_Q_QUEUED: | |
517 | /* | |
518 | * if the engine doesn't have a commit hook, | |
519 | * the io_u is really queued. if it does have such | |
520 | * a hook, it has to call io_u_queued() itself. | |
521 | */ | |
522 | if (td->io_ops->commit == NULL) | |
523 | io_u_queued(td, io_u); | |
524 | if (bytes_issued) | |
525 | *bytes_issued += io_u->xfer_buflen; | |
526 | break; | |
527 | case FIO_Q_BUSY: | |
528 | if (!from_verify) | |
529 | unlog_io_piece(td, io_u); | |
530 | requeue_io_u(td, &io_u); | |
a80cb54b | 531 | td_io_commit(td); |
e9d512d8 JA |
532 | break; |
533 | default: | |
b0775325 | 534 | assert(*ret < 0); |
e9d512d8 JA |
535 | td_verror(td, -(*ret), "td_io_queue"); |
536 | break; | |
537 | } | |
538 | ||
539 | if (break_on_this_error(td, ddir, ret)) | |
540 | return 1; | |
541 | ||
542 | return 0; | |
543 | } | |
544 | ||
e39c0676 | 545 | static inline bool io_in_polling(struct thread_data *td) |
82407585 RP |
546 | { |
547 | return !td->o.iodepth_batch_complete_min && | |
548 | !td->o.iodepth_batch_complete_max; | |
549 | } | |
39c1c323 | 550 | /* |
551 | * Unlinks files from thread data fio_file structure | |
552 | */ | |
553 | static int unlink_all_files(struct thread_data *td) | |
554 | { | |
555 | struct fio_file *f; | |
556 | unsigned int i; | |
557 | int ret = 0; | |
558 | ||
559 | for_each_file(td, f, i) { | |
560 | if (f->filetype != FIO_TYPE_FILE) | |
561 | continue; | |
562 | ret = td_io_unlink_file(td, f); | |
563 | if (ret) | |
564 | break; | |
565 | } | |
566 | ||
567 | if (ret) | |
568 | td_verror(td, ret, "unlink_all_files"); | |
569 | ||
570 | return ret; | |
571 | } | |
82407585 | 572 | |
997b5680 SW |
573 | /* |
574 | * Check if io_u will overlap an in-flight IO in the queue | |
575 | */ | |
c06379a6 | 576 | bool in_flight_overlap(struct io_u_queue *q, struct io_u *io_u) |
997b5680 SW |
577 | { |
578 | bool overlap; | |
579 | struct io_u *check_io_u; | |
580 | unsigned long long x1, x2, y1, y2; | |
581 | int i; | |
582 | ||
583 | x1 = io_u->offset; | |
584 | x2 = io_u->offset + io_u->buflen; | |
585 | overlap = false; | |
586 | io_u_qiter(q, check_io_u, i) { | |
587 | if (check_io_u->flags & IO_U_F_FLIGHT) { | |
588 | y1 = check_io_u->offset; | |
589 | y2 = check_io_u->offset + check_io_u->buflen; | |
590 | ||
591 | if (x1 < y2 && y1 < x2) { | |
592 | overlap = true; | |
5fff9543 | 593 | dprint(FD_IO, "in-flight overlap: %llu/%llu, %llu/%llu\n", |
997b5680 SW |
594 | x1, io_u->buflen, |
595 | y1, check_io_u->buflen); | |
596 | break; | |
597 | } | |
598 | } | |
599 | } | |
600 | ||
601 | return overlap; | |
602 | } | |
603 | ||
d3b07186 | 604 | static enum fio_q_status io_u_submit(struct thread_data *td, struct io_u *io_u) |
08b2b386 JA |
605 | { |
606 | /* | |
607 | * Check for overlap if the user asked us to, and we have | |
608 | * at least one IO in flight besides this one. | |
609 | */ | |
610 | if (td->o.serialize_overlap && td->cur_depth > 1 && | |
611 | in_flight_overlap(&td->io_u_all, io_u)) | |
612 | return FIO_Q_BUSY; | |
613 | ||
614 | return td_io_queue(td, io_u); | |
615 | } | |
616 | ||
2e1df07d JA |
617 | /* |
618 | * The main verify engine. Runs over the writes we previously submitted, | |
619 | * reads the blocks back in, and checks the crc/md5 of the data. | |
620 | */ | |
100f49f1 | 621 | static void do_verify(struct thread_data *td, uint64_t verify_bytes) |
2e1df07d JA |
622 | { |
623 | struct fio_file *f; | |
624 | struct io_u *io_u; | |
625 | int ret, min_events; | |
626 | unsigned int i; | |
627 | ||
628 | dprint(FD_VERIFY, "starting loop\n"); | |
629 | ||
630 | /* | |
631 | * sync io first and invalidate cache, to make sure we really | |
632 | * read from disk. | |
633 | */ | |
634 | for_each_file(td, f, i) { | |
635 | if (!fio_file_open(f)) | |
636 | continue; | |
637 | if (fio_io_sync(td, f)) | |
638 | break; | |
639 | if (file_invalidate_cache(td, f)) | |
640 | break; | |
641 | } | |
642 | ||
c97f1ad6 JA |
643 | check_update_rusage(td); |
644 | ||
2e1df07d JA |
645 | if (td->error) |
646 | return; | |
647 | ||
8f380de2 CJ |
648 | /* |
649 | * verify_state needs to be reset before verification | |
650 | * proceeds so that expected random seeds match actual | |
651 | * random seeds in headers. The main loop will reset | |
652 | * all random number generators if randrepeat is set. | |
653 | */ | |
654 | if (!td->o.rand_repeatable) | |
655 | td_fill_verify_state_seed(td); | |
656 | ||
2e1df07d JA |
657 | td_set_runstate(td, TD_VERIFYING); |
658 | ||
659 | io_u = NULL; | |
660 | while (!td->terminate) { | |
fbccf46c | 661 | enum fio_ddir ddir; |
e9d512d8 | 662 | int full; |
2e1df07d | 663 | |
8b6a404c | 664 | update_ts_cache(td); |
c97f1ad6 | 665 | check_update_rusage(td); |
2e1df07d | 666 | |
8b6a404c VF |
667 | if (runtime_exceeded(td, &td->ts_cache)) { |
668 | __update_ts_cache(td); | |
669 | if (runtime_exceeded(td, &td->ts_cache)) { | |
ebea2133 | 670 | fio_mark_td_terminate(td); |
2e1df07d JA |
671 | break; |
672 | } | |
673 | } | |
674 | ||
9e684a49 DE |
675 | if (flow_threshold_exceeded(td)) |
676 | continue; | |
677 | ||
44cbc6da JA |
678 | if (!td->o.experimental_verify) { |
679 | io_u = __get_io_u(td); | |
680 | if (!io_u) | |
681 | break; | |
2e1df07d | 682 | |
44cbc6da JA |
683 | if (get_next_verify(td, io_u)) { |
684 | put_io_u(td, io_u); | |
685 | break; | |
686 | } | |
2e1df07d | 687 | |
44cbc6da JA |
688 | if (td_io_prep(td, io_u)) { |
689 | put_io_u(td, io_u); | |
690 | break; | |
691 | } | |
692 | } else { | |
55312f9f | 693 | if (ddir_rw_sum(td->bytes_done) + td->o.rw_min_bs > verify_bytes) |
100f49f1 JA |
694 | break; |
695 | ||
bcd5abfa | 696 | while ((io_u = get_io_u(td)) != NULL) { |
059b61f2 | 697 | if (IS_ERR_OR_NULL(io_u)) { |
002fe734 JA |
698 | io_u = NULL; |
699 | ret = FIO_Q_BUSY; | |
700 | goto reap; | |
701 | } | |
702 | ||
bcd5abfa JA |
703 | /* |
704 | * We are only interested in the places where | |
705 | * we wrote or trimmed IOs. Turn those into | |
706 | * reads for verification purposes. | |
707 | */ | |
708 | if (io_u->ddir == DDIR_READ) { | |
709 | /* | |
710 | * Pretend we issued it for rwmix | |
711 | * accounting | |
712 | */ | |
713 | td->io_issues[DDIR_READ]++; | |
714 | put_io_u(td, io_u); | |
715 | continue; | |
716 | } else if (io_u->ddir == DDIR_TRIM) { | |
717 | io_u->ddir = DDIR_READ; | |
1651e431 | 718 | io_u_set(td, io_u, IO_U_F_TRIMMED); |
bcd5abfa JA |
719 | break; |
720 | } else if (io_u->ddir == DDIR_WRITE) { | |
721 | io_u->ddir = DDIR_READ; | |
5c5c33c1 | 722 | populate_verify_io_u(td, io_u); |
bcd5abfa JA |
723 | break; |
724 | } else { | |
725 | put_io_u(td, io_u); | |
726 | continue; | |
727 | } | |
728 | } | |
44cbc6da | 729 | |
bcd5abfa | 730 | if (!io_u) |
44cbc6da | 731 | break; |
2e1df07d JA |
732 | } |
733 | ||
ca09be4b JA |
734 | if (verify_state_should_stop(td, io_u)) { |
735 | put_io_u(td, io_u); | |
736 | break; | |
737 | } | |
738 | ||
2e1df07d JA |
739 | if (td->o.verify_async) |
740 | io_u->end_io = verify_io_u_async; | |
741 | else | |
742 | io_u->end_io = verify_io_u; | |
743 | ||
fbccf46c | 744 | ddir = io_u->ddir; |
10adc4a7 GG |
745 | if (!td->o.disable_slat) |
746 | fio_gettime(&io_u->start_time, NULL); | |
fbccf46c | 747 | |
08b2b386 | 748 | ret = io_u_submit(td, io_u); |
2e1df07d | 749 | |
e9d512d8 | 750 | if (io_queue_event(td, io_u, &ret, ddir, NULL, 1, NULL)) |
2e1df07d JA |
751 | break; |
752 | ||
753 | /* | |
754 | * if we can queue more, do so. but check if there are | |
755 | * completed io_u's first. Note that we can get BUSY even | |
756 | * without IO queued, if the system is resource starved. | |
757 | */ | |
002fe734 | 758 | reap: |
2e1df07d | 759 | full = queue_full(td) || (ret == FIO_Q_BUSY && td->cur_depth); |
82407585 | 760 | if (full || io_in_polling(td)) |
55312f9f | 761 | ret = wait_for_completions(td, NULL); |
2e1df07d | 762 | |
2e1df07d JA |
763 | if (ret < 0) |
764 | break; | |
765 | } | |
766 | ||
c97f1ad6 JA |
767 | check_update_rusage(td); |
768 | ||
2e1df07d JA |
769 | if (!td->error) { |
770 | min_events = td->cur_depth; | |
771 | ||
772 | if (min_events) | |
55312f9f | 773 | ret = io_u_queued_complete(td, min_events); |
2e1df07d JA |
774 | } else |
775 | cleanup_pending_aio(td); | |
776 | ||
777 | td_set_runstate(td, TD_RUNNING); | |
778 | ||
779 | dprint(FD_VERIFY, "exiting loop\n"); | |
780 | } | |
781 | ||
e39c0676 | 782 | static bool exceeds_number_ios(struct thread_data *td) |
3939fe85 JA |
783 | { |
784 | unsigned long long number_ios; | |
785 | ||
786 | if (!td->o.number_ios) | |
e39c0676 | 787 | return false; |
3939fe85 | 788 | |
cf8a46a7 | 789 | number_ios = ddir_rw_sum(td->io_blocks); |
3939fe85 JA |
790 | number_ios += td->io_u_queued + td->io_u_in_flight; |
791 | ||
cf8a46a7 | 792 | return number_ios >= (td->o.number_ios * td->loops); |
3939fe85 JA |
793 | } |
794 | ||
926098e0 | 795 | static bool io_bytes_exceeded(struct thread_data *td, uint64_t *this_bytes) |
f7078f7b | 796 | { |
77731b29 | 797 | unsigned long long bytes, limit; |
f7078f7b JA |
798 | |
799 | if (td_rw(td)) | |
926098e0 | 800 | bytes = this_bytes[DDIR_READ] + this_bytes[DDIR_WRITE]; |
f7078f7b | 801 | else if (td_write(td)) |
926098e0 | 802 | bytes = this_bytes[DDIR_WRITE]; |
6eaf09d6 | 803 | else if (td_read(td)) |
926098e0 | 804 | bytes = this_bytes[DDIR_READ]; |
f7078f7b | 805 | else |
926098e0 | 806 | bytes = this_bytes[DDIR_TRIM]; |
f7078f7b | 807 | |
5be9bf09 TK |
808 | if (td->o.io_size) |
809 | limit = td->o.io_size; | |
77731b29 JA |
810 | else |
811 | limit = td->o.size; | |
812 | ||
cf8a46a7 | 813 | limit *= td->loops; |
77731b29 | 814 | return bytes >= limit || exceeds_number_ios(td); |
f7078f7b JA |
815 | } |
816 | ||
926098e0 | 817 | static bool io_issue_bytes_exceeded(struct thread_data *td) |
e28dd2cf | 818 | { |
926098e0 JA |
819 | return io_bytes_exceeded(td, td->io_issue_bytes); |
820 | } | |
e28dd2cf | 821 | |
926098e0 JA |
822 | static bool io_complete_bytes_exceeded(struct thread_data *td) |
823 | { | |
824 | return io_bytes_exceeded(td, td->this_io_bytes); | |
e28dd2cf JE |
825 | } |
826 | ||
50a8ce86 D |
827 | /* |
828 | * used to calculate the next io time for rate control | |
829 | * | |
830 | */ | |
831 | static long long usec_for_io(struct thread_data *td, enum fio_ddir ddir) | |
832 | { | |
1a9bf814 | 833 | uint64_t bps = td->rate_bps[ddir]; |
50a8ce86 D |
834 | |
835 | assert(!(td->flags & TD_F_CHILD)); | |
ff6bb260 | 836 | |
6de65959 | 837 | if (td->o.rate_process == RATE_PROCESS_POISSON) { |
1a9bf814 JA |
838 | uint64_t val, iops; |
839 | ||
ff6bb260 | 840 | iops = bps / td->o.bs[ddir]; |
a77d139b | 841 | val = (int64_t) (1000000 / iops) * |
6c8611c7 | 842 | -logf(__rand_0_1(&td->poisson_state[ddir])); |
a77d139b | 843 | if (val) { |
98e51d7c JA |
844 | dprint(FD_RATE, "poisson rate iops=%llu, ddir=%d\n", |
845 | (unsigned long long) 1000000 / val, | |
846 | ddir); | |
a77d139b | 847 | } |
6c8611c7 JA |
848 | td->last_usec[ddir] += val; |
849 | return td->last_usec[ddir]; | |
ff6bb260 | 850 | } else if (bps) { |
1a9bf814 JA |
851 | uint64_t bytes = td->rate_io_issue_bytes[ddir]; |
852 | uint64_t secs = bytes / bps; | |
853 | uint64_t remainder = bytes % bps; | |
854 | ||
50a8ce86 | 855 | return remainder * 1000000 / bps + secs * 1000000; |
e7b24047 JA |
856 | } |
857 | ||
858 | return 0; | |
50a8ce86 D |
859 | } |
860 | ||
1a9bf814 JA |
861 | static void handle_thinktime(struct thread_data *td, enum fio_ddir ddir) |
862 | { | |
863 | unsigned long long b; | |
864 | uint64_t total; | |
865 | int left; | |
866 | ||
33f42c20 HQ |
867 | b = ddir_rw_sum(td->thinktime_blocks_counter); |
868 | if (b % td->o.thinktime_blocks || !b) | |
1a9bf814 JA |
869 | return; |
870 | ||
871 | io_u_quiesce(td); | |
872 | ||
873 | total = 0; | |
874 | if (td->o.thinktime_spin) | |
875 | total = usec_spin(td->o.thinktime_spin); | |
876 | ||
877 | left = td->o.thinktime - total; | |
878 | if (left) | |
879 | total += usec_sleep(td, left); | |
880 | ||
881 | /* | |
882 | * If we're ignoring thinktime for the rate, add the number of bytes | |
1aa39b0c JA |
883 | * we would have done while sleeping, minus one block to ensure we |
884 | * start issuing immediately after the sleep. | |
1a9bf814 | 885 | */ |
1aa39b0c | 886 | if (total && td->rate_bps[ddir] && td->o.rate_ign_think) { |
f2aedb17 | 887 | uint64_t missed = (td->rate_bps[ddir] * total) / 1000000ULL; |
62b858f6 JA |
888 | uint64_t bs = td->o.min_bs[ddir]; |
889 | uint64_t usperop = bs * 1000000ULL / td->rate_bps[ddir]; | |
f2aedb17 JA |
890 | uint64_t over; |
891 | ||
62b858f6 JA |
892 | if (usperop <= total) |
893 | over = bs; | |
f2aedb17 | 894 | else |
62b858f6 | 895 | over = (usperop - total) / usperop * -bs; |
f2aedb17 JA |
896 | |
897 | td->rate_io_issue_bytes[ddir] += (missed - over); | |
182cab24 JA |
898 | /* adjust for rate_process=poisson */ |
899 | td->last_usec[ddir] += total; | |
1aa39b0c | 900 | } |
1a9bf814 JA |
901 | } |
902 | ||
2e1df07d JA |
903 | /* |
904 | * Main IO worker function. It retrieves io_u's to process and queues | |
905 | * and reaps them, checking for rate and errors along the way. | |
100f49f1 JA |
906 | * |
907 | * Returns number of bytes written and trimmed. | |
2e1df07d | 908 | */ |
95b03be0 | 909 | static void do_io(struct thread_data *td, uint64_t *bytes_done) |
2e1df07d JA |
910 | { |
911 | unsigned int i; | |
912 | int ret = 0; | |
c2703bf3 | 913 | uint64_t total_bytes, bytes_issued = 0; |
70c68076 | 914 | |
95b03be0 JA |
915 | for (i = 0; i < DDIR_RWDIR_CNT; i++) |
916 | bytes_done[i] = td->bytes_done[i]; | |
2e1df07d JA |
917 | |
918 | if (in_ramp_time(td)) | |
919 | td_set_runstate(td, TD_RAMP); | |
920 | else | |
921 | td_set_runstate(td, TD_RUNNING); | |
922 | ||
3e260a46 JA |
923 | lat_target_init(td); |
924 | ||
1e564979 JE |
925 | total_bytes = td->o.size; |
926 | /* | |
5be9bf09 | 927 | * Allow random overwrite workloads to write up to io_size |
1e564979 JE |
928 | * before starting verification phase as 'size' doesn't apply. |
929 | */ | |
930 | if (td_write(td) && td_random(td) && td->o.norandommap) | |
5be9bf09 | 931 | total_bytes = max(total_bytes, (uint64_t) td->o.io_size); |
78a6469c JA |
932 | /* |
933 | * If verify_backlog is enabled, we'll run the verify in this | |
934 | * handler as well. For that case, we may need up to twice the | |
935 | * amount of bytes. | |
936 | */ | |
78a6469c JA |
937 | if (td->o.verify != VERIFY_NONE && |
938 | (td_write(td) && td->o.verify_backlog)) | |
c2703bf3 JA |
939 | total_bytes += td->o.size; |
940 | ||
82a90686 | 941 | /* In trimwrite mode, each byte is trimmed and then written, so |
0e4dd95c | 942 | * allow total_bytes to be twice as big */ |
82a90686 | 943 | if (td_trimwrite(td)) |
0e4dd95c DE |
944 | total_bytes += td->total_io_size; |
945 | ||
f7078f7b | 946 | while ((td->o.read_iolog_file && !flist_empty(&td->io_log_list)) || |
e28dd2cf | 947 | (!flist_empty(&td->trim_list)) || !io_issue_bytes_exceeded(td) || |
c04e4661 | 948 | td->o.time_based) { |
8b6a404c | 949 | struct timespec comp_time; |
2e1df07d | 950 | struct io_u *io_u; |
e9d512d8 | 951 | int full; |
2e1df07d JA |
952 | enum fio_ddir ddir; |
953 | ||
c97f1ad6 JA |
954 | check_update_rusage(td); |
955 | ||
7d7803fa | 956 | if (td->terminate || td->done) |
2e1df07d JA |
957 | break; |
958 | ||
8b6a404c | 959 | update_ts_cache(td); |
2e1df07d | 960 | |
8b6a404c VF |
961 | if (runtime_exceeded(td, &td->ts_cache)) { |
962 | __update_ts_cache(td); | |
963 | if (runtime_exceeded(td, &td->ts_cache)) { | |
ebea2133 | 964 | fio_mark_td_terminate(td); |
2e1df07d JA |
965 | break; |
966 | } | |
967 | } | |
968 | ||
9e684a49 DE |
969 | if (flow_threshold_exceeded(td)) |
970 | continue; | |
971 | ||
f1a32461 JA |
972 | /* |
973 | * Break if we exceeded the bytes. The exception is time | |
974 | * based runs, but we still need to break out of the loop | |
975 | * for those to run verification, if enabled. | |
98e7161c | 976 | * Jobs read from iolog do not use this stop condition. |
f1a32461 JA |
977 | */ |
978 | if (bytes_issued >= total_bytes && | |
98e7161c | 979 | !td->o.read_iolog_file && |
f1a32461 JA |
980 | (!td->o.time_based || |
981 | (td->o.time_based && td->o.verify != VERIFY_NONE))) | |
20876c53 JC |
982 | break; |
983 | ||
2e1df07d | 984 | io_u = get_io_u(td); |
002fe734 JA |
985 | if (IS_ERR_OR_NULL(io_u)) { |
986 | int err = PTR_ERR(io_u); | |
987 | ||
988 | io_u = NULL; | |
1a9bf814 | 989 | ddir = DDIR_INVAL; |
002fe734 JA |
990 | if (err == -EBUSY) { |
991 | ret = FIO_Q_BUSY; | |
992 | goto reap; | |
993 | } | |
3e260a46 JA |
994 | if (td->o.latency_target) |
995 | goto reap; | |
2e1df07d | 996 | break; |
3e260a46 | 997 | } |
2e1df07d | 998 | |
5c5c33c1 BVA |
999 | if (io_u->ddir == DDIR_WRITE && td->flags & TD_F_DO_VERIFY) |
1000 | populate_verify_io_u(td, io_u); | |
1001 | ||
2e1df07d JA |
1002 | ddir = io_u->ddir; |
1003 | ||
1004 | /* | |
82af2a7c JA |
1005 | * Add verification end_io handler if: |
1006 | * - Asked to verify (!td_rw(td)) | |
1007 | * - Or the io_u is from our verify list (mixed write/ver) | |
2e1df07d JA |
1008 | */ |
1009 | if (td->o.verify != VERIFY_NONE && io_u->ddir == DDIR_READ && | |
82af2a7c | 1010 | ((io_u->flags & IO_U_F_VER_LIST) || !td_rw(td))) { |
c4b6117b | 1011 | |
ca09be4b JA |
1012 | if (verify_state_should_stop(td, io_u)) { |
1013 | put_io_u(td, io_u); | |
1014 | break; | |
1015 | } | |
1016 | ||
2e1df07d JA |
1017 | if (td->o.verify_async) |
1018 | io_u->end_io = verify_io_u_async; | |
1019 | else | |
1020 | io_u->end_io = verify_io_u; | |
1021 | td_set_runstate(td, TD_VERIFYING); | |
1022 | } else if (in_ramp_time(td)) | |
1023 | td_set_runstate(td, TD_RAMP); | |
1024 | else | |
1025 | td_set_runstate(td, TD_RUNNING); | |
1026 | ||
9a50c5c5 | 1027 | /* |
f9401285 JA |
1028 | * Always log IO before it's issued, so we know the specific |
1029 | * order of it. The logged unit will track when the IO has | |
1030 | * completed. | |
9a50c5c5 | 1031 | */ |
c4b6117b PV |
1032 | if (td_write(td) && io_u->ddir == DDIR_WRITE && |
1033 | td->o.do_verify && | |
1034 | td->o.verify != VERIFY_NONE && | |
f9401285 | 1035 | !td->o.experimental_verify) |
c4b6117b PV |
1036 | log_io_piece(td, io_u); |
1037 | ||
a9da8ab2 | 1038 | if (td->o.io_submit_mode == IO_MODE_OFFLOAD) { |
5fff9543 | 1039 | const unsigned long long blen = io_u->xfer_buflen; |
a43f4461 | 1040 | const enum fio_ddir __ddir = acct_ddir(io_u); |
0c5df5f9 | 1041 | |
a9da8ab2 JA |
1042 | if (td->error) |
1043 | break; | |
0c5df5f9 | 1044 | |
26de50cf JA |
1045 | workqueue_enqueue(&td->io_wq, &io_u->work); |
1046 | ret = FIO_Q_QUEUED; | |
50a8ce86 | 1047 | |
a43f4461 JA |
1048 | if (ddir_rw(__ddir)) { |
1049 | td->io_issues[__ddir]++; | |
1050 | td->io_issue_bytes[__ddir] += blen; | |
1051 | td->rate_io_issue_bytes[__ddir] += blen; | |
0c5df5f9 JA |
1052 | } |
1053 | ||
50a8ce86 | 1054 | if (should_check_rate(td)) |
a43f4461 | 1055 | td->rate_next_io_time[__ddir] = usec_for_io(td, __ddir); |
50a8ce86 | 1056 | |
a9da8ab2 | 1057 | } else { |
08b2b386 | 1058 | ret = io_u_submit(td, io_u); |
2e1df07d | 1059 | |
50a8ce86 D |
1060 | if (should_check_rate(td)) |
1061 | td->rate_next_io_time[ddir] = usec_for_io(td, ddir); | |
1062 | ||
fd727d9d | 1063 | if (io_queue_event(td, io_u, &ret, ddir, &bytes_issued, 0, &comp_time)) |
a9da8ab2 | 1064 | break; |
2e1df07d | 1065 | |
a9da8ab2 JA |
1066 | /* |
1067 | * See if we need to complete some commands. Note that | |
1068 | * we can get BUSY even without IO queued, if the | |
1069 | * system is resource starved. | |
1070 | */ | |
3e260a46 | 1071 | reap: |
a9da8ab2 JA |
1072 | full = queue_full(td) || |
1073 | (ret == FIO_Q_BUSY && td->cur_depth); | |
82407585 | 1074 | if (full || io_in_polling(td)) |
a9da8ab2 JA |
1075 | ret = wait_for_completions(td, &comp_time); |
1076 | } | |
2e1df07d JA |
1077 | if (ret < 0) |
1078 | break; | |
33f42c20 HQ |
1079 | |
1080 | if (ddir_rw(ddir) && td->o.thinktime) | |
1081 | handle_thinktime(td, ddir); | |
1082 | ||
55312f9f | 1083 | if (!ddir_rw_sum(td->bytes_done) && |
9b87f09b | 1084 | !td_ioengine_flagged(td, FIO_NOIO)) |
2e1df07d JA |
1085 | continue; |
1086 | ||
55312f9f JA |
1087 | if (!in_ramp_time(td) && should_check_rate(td)) { |
1088 | if (check_min_rate(td, &comp_time)) { | |
f9cafb12 | 1089 | if (exitall_on_terminate || td->o.exitall_error) |
64402a8a | 1090 | fio_terminate_threads(td->groupid, td->o.exit_what); |
2e1df07d JA |
1091 | td_verror(td, EIO, "check_min_rate"); |
1092 | break; | |
1093 | } | |
1094 | } | |
3e260a46 JA |
1095 | if (!in_ramp_time(td) && td->o.latency_target) |
1096 | lat_target_check(td); | |
2e1df07d JA |
1097 | } |
1098 | ||
c97f1ad6 JA |
1099 | check_update_rusage(td); |
1100 | ||
2e1df07d | 1101 | if (td->trim_entries) |
4e0a8fa2 | 1102 | log_err("fio: %lu trim entries leaked?\n", td->trim_entries); |
2e1df07d JA |
1103 | |
1104 | if (td->o.fill_device && td->error == ENOSPC) { | |
1105 | td->error = 0; | |
ebea2133 | 1106 | fio_mark_td_terminate(td); |
2e1df07d JA |
1107 | } |
1108 | if (!td->error) { | |
1109 | struct fio_file *f; | |
1110 | ||
a9da8ab2 JA |
1111 | if (td->o.io_submit_mode == IO_MODE_OFFLOAD) { |
1112 | workqueue_flush(&td->io_wq); | |
1113 | i = 0; | |
1114 | } else | |
1115 | i = td->cur_depth; | |
1116 | ||
2e1df07d | 1117 | if (i) { |
55312f9f | 1118 | ret = io_u_queued_complete(td, i); |
2e1df07d JA |
1119 | if (td->o.fill_device && td->error == ENOSPC) |
1120 | td->error = 0; | |
1121 | } | |
1122 | ||
00615bfb | 1123 | if (should_fsync(td) && (td->o.end_fsync || td->o.fsync_on_close)) { |
2e1df07d JA |
1124 | td_set_runstate(td, TD_FSYNCING); |
1125 | ||
1126 | for_each_file(td, f, i) { | |
61ee0f86 | 1127 | if (!fio_file_fsync(td, f)) |
2e1df07d | 1128 | continue; |
61ee0f86 JA |
1129 | |
1130 | log_err("fio: end_fsync failed for file %s\n", | |
1131 | f->file_name); | |
2e1df07d JA |
1132 | } |
1133 | } | |
1134 | } else | |
1135 | cleanup_pending_aio(td); | |
1136 | ||
1137 | /* | |
1138 | * stop job if we failed doing any IO | |
1139 | */ | |
342f4be4 | 1140 | if (!ddir_rw_sum(td->this_io_bytes)) |
2e1df07d | 1141 | td->done = 1; |
100f49f1 | 1142 | |
95b03be0 JA |
1143 | for (i = 0; i < DDIR_RWDIR_CNT; i++) |
1144 | bytes_done[i] = td->bytes_done[i] - bytes_done[i]; | |
2e1df07d JA |
1145 | } |
1146 | ||
94a6e1bb JA |
1147 | static void free_file_completion_logging(struct thread_data *td) |
1148 | { | |
1149 | struct fio_file *f; | |
1150 | unsigned int i; | |
1151 | ||
1152 | for_each_file(td, f, i) { | |
1153 | if (!f->last_write_comp) | |
1154 | break; | |
1155 | sfree(f->last_write_comp); | |
1156 | } | |
1157 | } | |
1158 | ||
1159 | static int init_file_completion_logging(struct thread_data *td, | |
1160 | unsigned int depth) | |
1161 | { | |
1162 | struct fio_file *f; | |
1163 | unsigned int i; | |
1164 | ||
1165 | if (td->o.verify == VERIFY_NONE || !td->o.verify_state_save) | |
1166 | return 0; | |
1167 | ||
1168 | for_each_file(td, f, i) { | |
1169 | f->last_write_comp = scalloc(depth, sizeof(uint64_t)); | |
1170 | if (!f->last_write_comp) | |
1171 | goto cleanup; | |
1172 | } | |
1173 | ||
1174 | return 0; | |
1175 | ||
1176 | cleanup: | |
1177 | free_file_completion_logging(td); | |
1178 | log_err("fio: failed to alloc write comp data\n"); | |
1179 | return 1; | |
1180 | } | |
1181 | ||
2e1df07d JA |
1182 | static void cleanup_io_u(struct thread_data *td) |
1183 | { | |
2e1df07d JA |
1184 | struct io_u *io_u; |
1185 | ||
2ae0b204 | 1186 | while ((io_u = io_u_qpop(&td->io_u_freelist)) != NULL) { |
c73ed246 JA |
1187 | |
1188 | if (td->io_ops->io_u_free) | |
1189 | td->io_ops->io_u_free(td, io_u); | |
1190 | ||
3114b675 | 1191 | fio_memfree(io_u, sizeof(*io_u), td_offload_overlap(td)); |
2e1df07d JA |
1192 | } |
1193 | ||
1194 | free_io_mem(td); | |
2ae0b204 JA |
1195 | |
1196 | io_u_rexit(&td->io_u_requeues); | |
3114b675 VF |
1197 | io_u_qexit(&td->io_u_freelist, false); |
1198 | io_u_qexit(&td->io_u_all, td_offload_overlap(td)); | |
ca09be4b | 1199 | |
94a6e1bb | 1200 | free_file_completion_logging(td); |
2e1df07d JA |
1201 | } |
1202 | ||
1203 | static int init_io_u(struct thread_data *td) | |
1204 | { | |
1205 | struct io_u *io_u; | |
2e1df07d | 1206 | int cl_align, i, max_units; |
71e6e5a2 | 1207 | int err; |
2e1df07d JA |
1208 | |
1209 | max_units = td->o.iodepth; | |
59d8d0f5 | 1210 | |
2ae0b204 | 1211 | err = 0; |
34851ad5 | 1212 | err += !io_u_rinit(&td->io_u_requeues, td->o.iodepth); |
3114b675 VF |
1213 | err += !io_u_qinit(&td->io_u_freelist, td->o.iodepth, false); |
1214 | err += !io_u_qinit(&td->io_u_all, td->o.iodepth, td_offload_overlap(td)); | |
2ae0b204 JA |
1215 | |
1216 | if (err) { | |
1217 | log_err("fio: failed setting up IO queues\n"); | |
1218 | return 1; | |
1219 | } | |
1220 | ||
71e6e5a2 AK |
1221 | cl_align = os_cache_line_size(); |
1222 | ||
1223 | for (i = 0; i < max_units; i++) { | |
1224 | void *ptr; | |
1225 | ||
1226 | if (td->terminate) | |
1227 | return 1; | |
1228 | ||
3114b675 | 1229 | ptr = fio_memalign(cl_align, sizeof(*io_u), td_offload_overlap(td)); |
71e6e5a2 AK |
1230 | if (!ptr) { |
1231 | log_err("fio: unable to allocate aligned memory\n"); | |
ab43b8cb | 1232 | return 1; |
71e6e5a2 AK |
1233 | } |
1234 | ||
1235 | io_u = ptr; | |
1236 | memset(io_u, 0, sizeof(*io_u)); | |
1237 | INIT_FLIST_HEAD(&io_u->verify_list); | |
1238 | dprint(FD_MEM, "io_u alloc %p, index %u\n", io_u, i); | |
1239 | ||
1240 | io_u->index = i; | |
1241 | io_u->flags = IO_U_F_FREE; | |
1242 | io_u_qpush(&td->io_u_freelist, io_u); | |
1243 | ||
1244 | /* | |
1245 | * io_u never leaves this stack, used for iteration of all | |
1246 | * io_u buffers. | |
1247 | */ | |
1248 | io_u_qpush(&td->io_u_all, io_u); | |
1249 | ||
1250 | if (td->io_ops->io_u_init) { | |
1251 | int ret = td->io_ops->io_u_init(td, io_u); | |
1252 | ||
1253 | if (ret) { | |
1254 | log_err("fio: failed to init engine data: %d\n", ret); | |
1255 | return 1; | |
1256 | } | |
1257 | } | |
1258 | } | |
1259 | ||
1260 | init_io_u_buffers(td); | |
1261 | ||
1262 | if (init_file_completion_logging(td, max_units)) | |
1263 | return 1; | |
1264 | ||
1265 | return 0; | |
1266 | } | |
1267 | ||
1268 | int init_io_u_buffers(struct thread_data *td) | |
1269 | { | |
1270 | struct io_u *io_u; | |
1271 | unsigned long long max_bs, min_write; | |
1272 | int i, max_units; | |
1273 | int data_xfer = 1; | |
1274 | char *p; | |
1275 | ||
1276 | max_units = td->o.iodepth; | |
1277 | max_bs = td_max_bs(td); | |
1278 | min_write = td->o.min_bs[DDIR_WRITE]; | |
1279 | td->orig_buffer_size = (unsigned long long) max_bs | |
1280 | * (unsigned long long) max_units; | |
1281 | ||
1282 | if (td_ioengine_flagged(td, FIO_NOIO) || !(td_read(td) || td_write(td))) | |
1283 | data_xfer = 0; | |
1284 | ||
fd8a09b8 | 1285 | /* |
1286 | * if we may later need to do address alignment, then add any | |
1287 | * possible adjustment here so that we don't cause a buffer | |
1288 | * overflow later. this adjustment may be too much if we get | |
1289 | * lucky and the allocator gives us an aligned address. | |
1290 | */ | |
d01612f3 | 1291 | if (td->o.odirect || td->o.mem_align || td->o.oatomic || |
9b87f09b | 1292 | td_ioengine_flagged(td, FIO_RAWIO)) |
fd8a09b8 | 1293 | td->orig_buffer_size += page_mask + td->o.mem_align; |
1294 | ||
2e1df07d | 1295 | if (td->o.mem_type == MEM_SHMHUGE || td->o.mem_type == MEM_MMAPHUGE) { |
5fff9543 | 1296 | unsigned long long bs; |
2e1df07d JA |
1297 | |
1298 | bs = td->orig_buffer_size + td->o.hugepage_size - 1; | |
1299 | td->orig_buffer_size = bs & ~(td->o.hugepage_size - 1); | |
1300 | } | |
1301 | ||
1302 | if (td->orig_buffer_size != (size_t) td->orig_buffer_size) { | |
1303 | log_err("fio: IO memory too large. Reduce max_bs or iodepth\n"); | |
1304 | return 1; | |
1305 | } | |
1306 | ||
59d8d0f5 | 1307 | if (data_xfer && allocate_io_mem(td)) |
2e1df07d JA |
1308 | return 1; |
1309 | ||
d01612f3 | 1310 | if (td->o.odirect || td->o.mem_align || td->o.oatomic || |
9b87f09b | 1311 | td_ioengine_flagged(td, FIO_RAWIO)) |
248c9436 | 1312 | p = PTR_ALIGN(td->orig_buffer, page_mask) + td->o.mem_align; |
2e1df07d JA |
1313 | else |
1314 | p = td->orig_buffer; | |
1315 | ||
2e1df07d | 1316 | for (i = 0; i < max_units; i++) { |
71e6e5a2 | 1317 | io_u = td->io_u_all.io_us[i]; |
2e1df07d JA |
1318 | dprint(FD_MEM, "io_u alloc %p, index %u\n", io_u, i); |
1319 | ||
59d8d0f5 | 1320 | if (data_xfer) { |
2e1df07d JA |
1321 | io_u->buf = p; |
1322 | dprint(FD_MEM, "io_u %p, mem %p\n", io_u, io_u->buf); | |
1323 | ||
1324 | if (td_write(td)) | |
9c42684e | 1325 | io_u_fill_buffer(td, io_u, min_write, max_bs); |
2e1df07d JA |
1326 | if (td_write(td) && td->o.verify_pattern_bytes) { |
1327 | /* | |
1328 | * Fill the buffer with the pattern if we are | |
1329 | * going to be doing writes. | |
1330 | */ | |
ce35b1ec | 1331 | fill_verify_pattern(td, io_u->buf, max_bs, io_u, 0, 0); |
2e1df07d JA |
1332 | } |
1333 | } | |
2e1df07d JA |
1334 | p += max_bs; |
1335 | } | |
1336 | ||
1337 | return 0; | |
1338 | } | |
1339 | ||
bd4d9bdc TK |
1340 | /* |
1341 | * This function is Linux specific. | |
1342 | * FIO_HAVE_IOSCHED_SWITCH enabled currently means it's Linux. | |
1343 | */ | |
2e1df07d JA |
1344 | static int switch_ioscheduler(struct thread_data *td) |
1345 | { | |
db815fb3 | 1346 | #ifdef FIO_HAVE_IOSCHED_SWITCH |
178ef27a | 1347 | char tmp[256], tmp2[128], *p; |
2e1df07d JA |
1348 | FILE *f; |
1349 | int ret; | |
1350 | ||
9b87f09b | 1351 | if (td_ioengine_flagged(td, FIO_DISKLESSIO)) |
2e1df07d JA |
1352 | return 0; |
1353 | ||
bd4d9bdc TK |
1354 | assert(td->files && td->files[0]); |
1355 | sprintf(tmp, "%s/queue/scheduler", td->files[0]->du->sysfs_root); | |
2e1df07d JA |
1356 | |
1357 | f = fopen(tmp, "r+"); | |
1358 | if (!f) { | |
1359 | if (errno == ENOENT) { | |
1360 | log_err("fio: os or kernel doesn't support IO scheduler" | |
1361 | " switching\n"); | |
1362 | return 0; | |
1363 | } | |
1364 | td_verror(td, errno, "fopen iosched"); | |
1365 | return 1; | |
1366 | } | |
1367 | ||
1368 | /* | |
1369 | * Set io scheduler. | |
1370 | */ | |
1371 | ret = fwrite(td->o.ioscheduler, strlen(td->o.ioscheduler), 1, f); | |
1372 | if (ferror(f) || ret != 1) { | |
1373 | td_verror(td, errno, "fwrite"); | |
1374 | fclose(f); | |
1375 | return 1; | |
1376 | } | |
1377 | ||
1378 | rewind(f); | |
1379 | ||
1380 | /* | |
1381 | * Read back and check that the selected scheduler is now the default. | |
1382 | */ | |
0e23d456 | 1383 | ret = fread(tmp, 1, sizeof(tmp) - 1, f); |
2e1df07d JA |
1384 | if (ferror(f) || ret < 0) { |
1385 | td_verror(td, errno, "fread"); | |
1386 | fclose(f); | |
1387 | return 1; | |
1388 | } | |
0e23d456 | 1389 | tmp[ret] = '\0'; |
b44b9e45 | 1390 | /* |
178ef27a BVA |
1391 | * either a list of io schedulers or "none\n" is expected. Strip the |
1392 | * trailing newline. | |
b44b9e45 | 1393 | */ |
178ef27a BVA |
1394 | p = tmp; |
1395 | strsep(&p, "\n"); | |
49c6f33d | 1396 | |
67f28a8c TK |
1397 | /* |
1398 | * Write to "none" entry doesn't fail, so check the result here. | |
1399 | */ | |
1400 | if (!strcmp(tmp, "none")) { | |
1401 | log_err("fio: io scheduler is not tunable\n"); | |
1402 | fclose(f); | |
1403 | return 0; | |
1404 | } | |
2e1df07d JA |
1405 | |
1406 | sprintf(tmp2, "[%s]", td->o.ioscheduler); | |
1407 | if (!strstr(tmp, tmp2)) { | |
1408 | log_err("fio: io scheduler %s not found\n", td->o.ioscheduler); | |
1409 | td_verror(td, EINVAL, "iosched_switch"); | |
1410 | fclose(f); | |
1411 | return 1; | |
1412 | } | |
1413 | ||
1414 | fclose(f); | |
1415 | return 0; | |
db815fb3 TK |
1416 | #else |
1417 | return 0; | |
1418 | #endif | |
2e1df07d JA |
1419 | } |
1420 | ||
e39c0676 | 1421 | static bool keep_running(struct thread_data *td) |
2e1df07d | 1422 | { |
77731b29 JA |
1423 | unsigned long long limit; |
1424 | ||
2e1df07d | 1425 | if (td->done) |
e39c0676 | 1426 | return false; |
98413be7 SW |
1427 | if (td->terminate) |
1428 | return false; | |
2e1df07d | 1429 | if (td->o.time_based) |
e39c0676 | 1430 | return true; |
2e1df07d JA |
1431 | if (td->o.loops) { |
1432 | td->o.loops--; | |
e39c0676 | 1433 | return true; |
2e1df07d | 1434 | } |
3939fe85 | 1435 | if (exceeds_number_ios(td)) |
e39c0676 | 1436 | return false; |
26251d8d | 1437 | |
5be9bf09 TK |
1438 | if (td->o.io_size) |
1439 | limit = td->o.io_size; | |
77731b29 JA |
1440 | else |
1441 | limit = td->o.size; | |
1442 | ||
1443 | if (limit != -1ULL && ddir_rw_sum(td->io_bytes) < limit) { | |
5bd5f71a JA |
1444 | uint64_t diff; |
1445 | ||
1446 | /* | |
589d5d4f | 1447 | * If the difference is less than the maximum IO size, we |
5bd5f71a JA |
1448 | * are done. |
1449 | */ | |
77731b29 | 1450 | diff = limit - ddir_rw_sum(td->io_bytes); |
74f4b020 | 1451 | if (diff < td_max_bs(td)) |
e39c0676 | 1452 | return false; |
5bd5f71a | 1453 | |
5be9bf09 | 1454 | if (fio_files_done(td) && !td->o.io_size) |
e39c0676 | 1455 | return false; |
002fe734 | 1456 | |
e39c0676 | 1457 | return true; |
5bd5f71a | 1458 | } |
2e1df07d | 1459 | |
e39c0676 | 1460 | return false; |
2e1df07d JA |
1461 | } |
1462 | ||
b7e4a53c BVA |
1463 | static int exec_string(struct thread_options *o, const char *string, |
1464 | const char *mode) | |
2e1df07d | 1465 | { |
f491a907 | 1466 | int ret; |
2e1df07d JA |
1467 | char *str; |
1468 | ||
b7e4a53c BVA |
1469 | if (asprintf(&str, "%s > %s.%s.txt 2>&1", string, o->name, mode) < 0) |
1470 | return -1; | |
2e1df07d | 1471 | |
b7e4a53c BVA |
1472 | log_info("%s : Saving output of %s in %s.%s.txt\n", o->name, mode, |
1473 | o->name, mode); | |
2e1df07d JA |
1474 | ret = system(str); |
1475 | if (ret == -1) | |
1476 | log_err("fio: exec of cmd <%s> failed\n", str); | |
1477 | ||
1478 | free(str); | |
1479 | return ret; | |
1480 | } | |
1481 | ||
62167762 JC |
1482 | /* |
1483 | * Dry run to compute correct state of numberio for verification. | |
1484 | */ | |
1485 | static uint64_t do_dry_run(struct thread_data *td) | |
1486 | { | |
62167762 JC |
1487 | td_set_runstate(td, TD_RUNNING); |
1488 | ||
1489 | while ((td->o.read_iolog_file && !flist_empty(&td->io_log_list)) || | |
e28dd2cf | 1490 | (!flist_empty(&td->trim_list)) || !io_complete_bytes_exceeded(td)) { |
62167762 JC |
1491 | struct io_u *io_u; |
1492 | int ret; | |
1493 | ||
1494 | if (td->terminate || td->done) | |
1495 | break; | |
1496 | ||
1497 | io_u = get_io_u(td); | |
99955d3d | 1498 | if (IS_ERR_OR_NULL(io_u)) |
62167762 JC |
1499 | break; |
1500 | ||
1651e431 | 1501 | io_u_set(td, io_u, IO_U_F_FLIGHT); |
62167762 JC |
1502 | io_u->error = 0; |
1503 | io_u->resid = 0; | |
1504 | if (ddir_rw(acct_ddir(io_u))) | |
1505 | td->io_issues[acct_ddir(io_u)]++; | |
1506 | if (ddir_rw(io_u->ddir)) { | |
1507 | io_u_mark_depth(td, 1); | |
1508 | td->ts.total_io_u[io_u->ddir]++; | |
1509 | } | |
1510 | ||
2e63e96b PV |
1511 | if (td_write(td) && io_u->ddir == DDIR_WRITE && |
1512 | td->o.do_verify && | |
1513 | td->o.verify != VERIFY_NONE && | |
1514 | !td->o.experimental_verify) | |
1515 | log_io_piece(td, io_u); | |
1516 | ||
55312f9f | 1517 | ret = io_u_sync_complete(td, io_u); |
62167762 JC |
1518 | (void) ret; |
1519 | } | |
1520 | ||
55312f9f | 1521 | return td->bytes_done[DDIR_WRITE] + td->bytes_done[DDIR_TRIM]; |
62167762 JC |
1522 | } |
1523 | ||
24660963 JA |
1524 | struct fork_data { |
1525 | struct thread_data *td; | |
1526 | struct sk_out *sk_out; | |
1527 | }; | |
1528 | ||
2e1df07d JA |
1529 | /* |
1530 | * Entry point for the thread based jobs. The process based jobs end up | |
1531 | * here as well, after a little setup. | |
1532 | */ | |
1533 | static void *thread_main(void *data) | |
1534 | { | |
24660963 | 1535 | struct fork_data *fd = data; |
95603b74 | 1536 | unsigned long long elapsed_us[DDIR_RWDIR_CNT] = { 0, }; |
24660963 | 1537 | struct thread_data *td = fd->td; |
4896473e | 1538 | struct thread_options *o = &td->o; |
24660963 | 1539 | struct sk_out *sk_out = fd->sk_out; |
48366f3a | 1540 | uint64_t bytes_done[DDIR_RWDIR_CNT]; |
ca0122d8 | 1541 | int deadlock_loop_cnt; |
ff3aa922 | 1542 | bool clear_state; |
a38c70a8 | 1543 | int res, ret; |
2e1df07d | 1544 | |
24660963 JA |
1545 | sk_out_assign(sk_out); |
1546 | free(fd); | |
1547 | ||
4896473e | 1548 | if (!o->use_thread) { |
2e1df07d JA |
1549 | setsid(); |
1550 | td->pid = getpid(); | |
1551 | } else | |
1552 | td->pid = gettid(); | |
1553 | ||
a0eba820 | 1554 | fio_local_clock_init(); |
5d879392 | 1555 | |
2e1df07d JA |
1556 | dprint(FD_PROCESS, "jobs pid=%d started\n", (int) td->pid); |
1557 | ||
122c7725 JA |
1558 | if (is_backend) |
1559 | fio_server_send_start(td); | |
1560 | ||
2e1df07d JA |
1561 | INIT_FLIST_HEAD(&td->io_log_list); |
1562 | INIT_FLIST_HEAD(&td->io_hist_list); | |
1563 | INIT_FLIST_HEAD(&td->verify_list); | |
1564 | INIT_FLIST_HEAD(&td->trim_list); | |
2e1df07d JA |
1565 | td->io_hist_tree = RB_ROOT; |
1566 | ||
34febb23 | 1567 | ret = mutex_cond_init_pshared(&td->io_u_lock, &td->free_cond); |
f9e5b5ee | 1568 | if (ret) { |
34febb23 | 1569 | td_verror(td, ret, "mutex_cond_init_pshared"); |
f9e5b5ee JK |
1570 | goto err; |
1571 | } | |
34febb23 | 1572 | ret = cond_init_pshared(&td->verify_cond); |
f9e5b5ee | 1573 | if (ret) { |
34febb23 | 1574 | td_verror(td, ret, "mutex_cond_pshared"); |
f9e5b5ee JK |
1575 | goto err; |
1576 | } | |
2e1df07d JA |
1577 | |
1578 | td_set_runstate(td, TD_INITIALIZED); | |
971caeb1 BVA |
1579 | dprint(FD_MUTEX, "up startup_sem\n"); |
1580 | fio_sem_up(startup_sem); | |
1581 | dprint(FD_MUTEX, "wait on td->sem\n"); | |
1582 | fio_sem_down(td->sem); | |
1583 | dprint(FD_MUTEX, "done waiting on td->sem\n"); | |
2e1df07d | 1584 | |
2e1df07d JA |
1585 | /* |
1586 | * A new gid requires privilege, so we need to do this before setting | |
1587 | * the uid. | |
1588 | */ | |
4896473e | 1589 | if (o->gid != -1U && setgid(o->gid)) { |
2e1df07d JA |
1590 | td_verror(td, errno, "setgid"); |
1591 | goto err; | |
1592 | } | |
4896473e | 1593 | if (o->uid != -1U && setuid(o->uid)) { |
2e1df07d JA |
1594 | td_verror(td, errno, "setuid"); |
1595 | goto err; | |
1596 | } | |
1597 | ||
07776127 JA |
1598 | td_zone_gen_index(td); |
1599 | ||
2354d810 JA |
1600 | /* |
1601 | * Do this early, we don't want the compress threads to be limited | |
1602 | * to the same CPUs as the IO workers. So do this before we set | |
1603 | * any potential CPU affinity | |
1604 | */ | |
1605 | if (iolog_compress_init(td, sk_out)) | |
1606 | goto err; | |
1607 | ||
2e1df07d JA |
1608 | /* |
1609 | * If we have a gettimeofday() thread, make sure we exclude that | |
1610 | * thread from this job | |
1611 | */ | |
4896473e JA |
1612 | if (o->gtod_cpu) |
1613 | fio_cpu_clear(&o->cpumask, o->gtod_cpu); | |
2e1df07d JA |
1614 | |
1615 | /* | |
1616 | * Set affinity first, in case it has an impact on the memory | |
1617 | * allocations. | |
1618 | */ | |
b2a9e649 | 1619 | if (fio_option_is_set(o, cpumask)) { |
c2acfbac | 1620 | if (o->cpus_allowed_policy == FIO_CPUS_SPLIT) { |
30cb4c65 | 1621 | ret = fio_cpus_split(&o->cpumask, td->thread_number - 1); |
c2acfbac JA |
1622 | if (!ret) { |
1623 | log_err("fio: no CPUs set\n"); | |
1624 | log_err("fio: Try increasing number of available CPUs\n"); | |
1625 | td_verror(td, EINVAL, "cpus_split"); | |
1626 | goto err; | |
1627 | } | |
1628 | } | |
28727df7 JA |
1629 | ret = fio_setaffinity(td->pid, o->cpumask); |
1630 | if (ret == -1) { | |
4896473e JA |
1631 | td_verror(td, errno, "cpu_set_affinity"); |
1632 | goto err; | |
1633 | } | |
2e1df07d JA |
1634 | } |
1635 | ||
67bf9823 | 1636 | #ifdef CONFIG_LIBNUMA |
d0b937ed | 1637 | /* numa node setup */ |
b2a9e649 JA |
1638 | if (fio_option_is_set(o, numa_cpunodes) || |
1639 | fio_option_is_set(o, numa_memnodes)) { | |
43522848 | 1640 | struct bitmask *mask; |
d0b937ed YR |
1641 | |
1642 | if (numa_available() < 0) { | |
1643 | td_verror(td, errno, "Does not support NUMA API\n"); | |
1644 | goto err; | |
1645 | } | |
1646 | ||
b2a9e649 | 1647 | if (fio_option_is_set(o, numa_cpunodes)) { |
43522848 DG |
1648 | mask = numa_parse_nodestring(o->numa_cpunodes); |
1649 | ret = numa_run_on_node_mask(mask); | |
1650 | numa_free_nodemask(mask); | |
d0b937ed YR |
1651 | if (ret == -1) { |
1652 | td_verror(td, errno, \ | |
1653 | "numa_run_on_node_mask failed\n"); | |
1654 | goto err; | |
1655 | } | |
1656 | } | |
1657 | ||
b2a9e649 | 1658 | if (fio_option_is_set(o, numa_memnodes)) { |
43522848 DG |
1659 | mask = NULL; |
1660 | if (o->numa_memnodes) | |
1661 | mask = numa_parse_nodestring(o->numa_memnodes); | |
1662 | ||
4896473e | 1663 | switch (o->numa_mem_mode) { |
d0b937ed | 1664 | case MPOL_INTERLEAVE: |
43522848 | 1665 | numa_set_interleave_mask(mask); |
d0b937ed YR |
1666 | break; |
1667 | case MPOL_BIND: | |
43522848 | 1668 | numa_set_membind(mask); |
d0b937ed YR |
1669 | break; |
1670 | case MPOL_LOCAL: | |
1671 | numa_set_localalloc(); | |
1672 | break; | |
1673 | case MPOL_PREFERRED: | |
4896473e | 1674 | numa_set_preferred(o->numa_mem_prefer_node); |
d0b937ed YR |
1675 | break; |
1676 | case MPOL_DEFAULT: | |
1677 | default: | |
1678 | break; | |
1679 | } | |
1680 | ||
43522848 DG |
1681 | if (mask) |
1682 | numa_free_nodemask(mask); | |
1683 | ||
d0b937ed YR |
1684 | } |
1685 | } | |
1686 | #endif | |
1687 | ||
9a3f1100 JA |
1688 | if (fio_pin_memory(td)) |
1689 | goto err; | |
1690 | ||
2e1df07d JA |
1691 | /* |
1692 | * May alter parameters that init_io_u() will use, so we need to | |
1693 | * do this first. | |
1694 | */ | |
b153f94a | 1695 | if (!init_iolog(td)) |
2e1df07d JA |
1696 | goto err; |
1697 | ||
08dc3bd5 JA |
1698 | if (td_io_init(td)) |
1699 | goto err; | |
1700 | ||
2e1df07d JA |
1701 | if (init_io_u(td)) |
1702 | goto err; | |
1703 | ||
2041bd34 JA |
1704 | if (td->io_ops->post_init && td->io_ops->post_init(td)) |
1705 | goto err; | |
1706 | ||
4896473e | 1707 | if (o->verify_async && verify_async_init(td)) |
2e1df07d JA |
1708 | goto err; |
1709 | ||
27d74836 JA |
1710 | if (fio_option_is_set(o, ioprio) || |
1711 | fio_option_is_set(o, ioprio_class)) { | |
28727df7 JA |
1712 | ret = ioprio_set(IOPRIO_WHO_PROCESS, 0, o->ioprio_class, o->ioprio); |
1713 | if (ret == -1) { | |
2e1df07d JA |
1714 | td_verror(td, errno, "ioprio_set"); |
1715 | goto err; | |
1716 | } | |
1717 | } | |
1718 | ||
4896473e | 1719 | if (o->cgroup && cgroup_setup(td, cgroup_list, &cgroup_mnt)) |
2e1df07d JA |
1720 | goto err; |
1721 | ||
649c10c5 | 1722 | errno = 0; |
4896473e | 1723 | if (nice(o->nice) == -1 && errno != 0) { |
2e1df07d JA |
1724 | td_verror(td, errno, "nice"); |
1725 | goto err; | |
1726 | } | |
1727 | ||
4896473e | 1728 | if (o->ioscheduler && switch_ioscheduler(td)) |
2e1df07d JA |
1729 | goto err; |
1730 | ||
4896473e | 1731 | if (!o->create_serialize && setup_files(td)) |
2e1df07d JA |
1732 | goto err; |
1733 | ||
4d832322 | 1734 | if (!init_random_map(td)) |
2e1df07d JA |
1735 | goto err; |
1736 | ||
0518f09f | 1737 | if (o->exec_prerun && exec_string(o, o->exec_prerun, "prerun")) |
4896473e | 1738 | goto err; |
2e1df07d | 1739 | |
c139f3b4 JA |
1740 | if (o->pre_read && !pre_read_files(td)) |
1741 | goto err; | |
2e1df07d | 1742 | |
dc5bfbb2 JA |
1743 | fio_verify_init(td); |
1744 | ||
24660963 | 1745 | if (rate_submit_init(td, sk_out)) |
a9da8ab2 JA |
1746 | goto err; |
1747 | ||
33f42c20 HQ |
1748 | if (td->o.thinktime_blocks_type == THINKTIME_BLOCKS_TYPE_COMPLETE) |
1749 | td->thinktime_blocks_counter = td->io_blocks; | |
1750 | else | |
1751 | td->thinktime_blocks_counter = td->io_issues; | |
1752 | ||
3aea75b1 | 1753 | set_epoch_time(td, o->log_unix_epoch); |
44404c5a | 1754 | fio_getrusage(&td->ru_start); |
ac28d905 JA |
1755 | memcpy(&td->bw_sample_time, &td->epoch, sizeof(td->epoch)); |
1756 | memcpy(&td->iops_sample_time, &td->epoch, sizeof(td->epoch)); | |
16e56d25 | 1757 | memcpy(&td->ss.prev_time, &td->epoch, sizeof(td->epoch)); |
ac28d905 JA |
1758 | |
1759 | if (o->ratemin[DDIR_READ] || o->ratemin[DDIR_WRITE] || | |
1760 | o->ratemin[DDIR_TRIM]) { | |
1761 | memcpy(&td->lastrate[DDIR_READ], &td->bw_sample_time, | |
1762 | sizeof(td->bw_sample_time)); | |
1763 | memcpy(&td->lastrate[DDIR_WRITE], &td->bw_sample_time, | |
1764 | sizeof(td->bw_sample_time)); | |
1765 | memcpy(&td->lastrate[DDIR_TRIM], &td->bw_sample_time, | |
1766 | sizeof(td->bw_sample_time)); | |
1767 | } | |
1768 | ||
48366f3a | 1769 | memset(bytes_done, 0, sizeof(bytes_done)); |
c16035aa | 1770 | clear_state = false; |
48366f3a | 1771 | |
2e1df07d | 1772 | while (keep_running(td)) { |
100f49f1 JA |
1773 | uint64_t verify_bytes; |
1774 | ||
2e1df07d | 1775 | fio_gettime(&td->start, NULL); |
8b6a404c | 1776 | memcpy(&td->ts_cache, &td->start, sizeof(td->start)); |
2e1df07d | 1777 | |
39c1c323 | 1778 | if (clear_state) { |
ac28d905 | 1779 | clear_io_state(td, 0); |
2e1df07d | 1780 | |
39c1c323 | 1781 | if (o->unlink_each_loop && unlink_all_files(td)) |
1782 | break; | |
1783 | } | |
1784 | ||
2e1df07d JA |
1785 | prune_io_piece_log(td); |
1786 | ||
dcf9844e | 1787 | if (td->o.verify_only && td_write(td)) |
62167762 | 1788 | verify_bytes = do_dry_run(td); |
095196b1 | 1789 | else { |
95b03be0 JA |
1790 | do_io(td, bytes_done); |
1791 | ||
1792 | if (!ddir_rw_sum(bytes_done)) { | |
095196b1 | 1793 | fio_mark_td_terminate(td); |
95b03be0 JA |
1794 | verify_bytes = 0; |
1795 | } else { | |
1796 | verify_bytes = bytes_done[DDIR_WRITE] + | |
1797 | bytes_done[DDIR_TRIM]; | |
1798 | } | |
095196b1 | 1799 | } |
2e1df07d | 1800 | |
49f0b8e0 JA |
1801 | /* |
1802 | * If we took too long to shut down, the main thread could | |
1803 | * already consider us reaped/exited. If that happens, break | |
1804 | * out and clean up. | |
1805 | */ | |
1806 | if (td->runstate >= TD_EXITED) | |
1807 | break; | |
1808 | ||
c16035aa | 1809 | clear_state = true; |
2e1df07d | 1810 | |
40c666c8 JA |
1811 | /* |
1812 | * Make sure we've successfully updated the rusage stats | |
1813 | * before waiting on the stat mutex. Otherwise we could have | |
1814 | * the stat thread holding stat mutex and waiting for | |
1815 | * the rusage_sem, which would never get upped because | |
1816 | * this thread is waiting for the stat mutex. | |
1817 | */ | |
ca0122d8 | 1818 | deadlock_loop_cnt = 0; |
39d13e67 JA |
1819 | do { |
1820 | check_update_rusage(td); | |
971caeb1 | 1821 | if (!fio_sem_down_trylock(stat_sem)) |
39d13e67 JA |
1822 | break; |
1823 | usleep(1000); | |
ca0122d8 | 1824 | if (deadlock_loop_cnt++ > 5000) { |
971caeb1 | 1825 | log_err("fio seems to be stuck grabbing stat_sem, forcibly exiting\n"); |
43f248c4 | 1826 | td->error = EDEADLK; |
ca0122d8 TT |
1827 | goto err; |
1828 | } | |
39d13e67 | 1829 | } while (1); |
40c666c8 | 1830 | |
95603b74 BF |
1831 | if (td_read(td) && td->io_bytes[DDIR_READ]) |
1832 | update_runtime(td, elapsed_us, DDIR_READ); | |
1833 | if (td_write(td) && td->io_bytes[DDIR_WRITE]) | |
1834 | update_runtime(td, elapsed_us, DDIR_WRITE); | |
1835 | if (td_trim(td) && td->io_bytes[DDIR_TRIM]) | |
1836 | update_runtime(td, elapsed_us, DDIR_TRIM); | |
e5437a07 | 1837 | fio_gettime(&td->start, NULL); |
971caeb1 | 1838 | fio_sem_up(stat_sem); |
2e1df07d JA |
1839 | |
1840 | if (td->error || td->terminate) | |
1841 | break; | |
1842 | ||
4896473e JA |
1843 | if (!o->do_verify || |
1844 | o->verify == VERIFY_NONE || | |
9b87f09b | 1845 | td_ioengine_flagged(td, FIO_UNIDIR)) |
2e1df07d JA |
1846 | continue; |
1847 | ||
ac28d905 | 1848 | clear_io_state(td, 0); |
2e1df07d JA |
1849 | |
1850 | fio_gettime(&td->start, NULL); | |
1851 | ||
100f49f1 | 1852 | do_verify(td, verify_bytes); |
2e1df07d | 1853 | |
40c666c8 JA |
1854 | /* |
1855 | * See comment further up for why this is done here. | |
1856 | */ | |
1857 | check_update_rusage(td); | |
1858 | ||
971caeb1 | 1859 | fio_sem_down(stat_sem); |
95603b74 | 1860 | update_runtime(td, elapsed_us, DDIR_READ); |
e5437a07 | 1861 | fio_gettime(&td->start, NULL); |
971caeb1 | 1862 | fio_sem_up(stat_sem); |
2e1df07d JA |
1863 | |
1864 | if (td->error || td->terminate) | |
1865 | break; | |
1866 | } | |
1867 | ||
817ae977 VF |
1868 | /* |
1869 | * Acquire this lock if we were doing overlap checking in | |
1870 | * offload mode so that we don't clean up this job while | |
1871 | * another thread is checking its io_u's for overlap | |
1872 | */ | |
a38c70a8 BVA |
1873 | if (td_offload_overlap(td)) { |
1874 | int res = pthread_mutex_lock(&overlap_check); | |
1875 | assert(res == 0); | |
1876 | } | |
2cea0b4c | 1877 | td_set_runstate(td, TD_FINISHING); |
a38c70a8 BVA |
1878 | if (td_offload_overlap(td)) { |
1879 | res = pthread_mutex_unlock(&overlap_check); | |
1880 | assert(res == 0); | |
1881 | } | |
2cea0b4c | 1882 | |
2e1df07d | 1883 | update_rusage_stat(td); |
2e1df07d | 1884 | td->ts.total_run_time = mtime_since_now(&td->epoch); |
dde39f8c AD |
1885 | for_each_rw_ddir(ddir) { |
1886 | td->ts.io_bytes[ddir] = td->io_bytes[ddir]; | |
1887 | } | |
2e1df07d | 1888 | |
ca09be4b | 1889 | if (td->o.verify_state_save && !(td->flags & TD_F_VSTATE_SAVED) && |
d264264a JA |
1890 | (td->o.verify != VERIFY_NONE && td_write(td))) |
1891 | verify_save_state(td->thread_number); | |
ca09be4b | 1892 | |
9a3f1100 JA |
1893 | fio_unpin_memory(td); |
1894 | ||
a47591e4 | 1895 | td_writeout_logs(td, true); |
2e1df07d | 1896 | |
155f2f02 | 1897 | iolog_compress_exit(td); |
103b174e | 1898 | rate_submit_exit(td); |
aee2ab67 | 1899 | |
4896473e | 1900 | if (o->exec_postrun) |
0518f09f | 1901 | exec_string(o, o->exec_postrun, "postrun"); |
2e1df07d | 1902 | |
f9cafb12 | 1903 | if (exitall_on_terminate || (o->exitall_error && td->error)) |
64402a8a | 1904 | fio_terminate_threads(td->groupid, td->o.exit_what); |
2e1df07d JA |
1905 | |
1906 | err: | |
1907 | if (td->error) | |
1908 | log_info("fio: pid=%d, err=%d/%s\n", (int) td->pid, td->error, | |
1909 | td->verror); | |
1910 | ||
4896473e | 1911 | if (o->verify_async) |
2e1df07d JA |
1912 | verify_async_exit(td); |
1913 | ||
1914 | close_and_free_files(td); | |
2e1df07d | 1915 | cleanup_io_u(td); |
32dbca2c | 1916 | close_ioengine(td); |
c3dc516a | 1917 | cgroup_shutdown(td, cgroup_mnt); |
ca09be4b | 1918 | verify_free_state(td); |
07776127 | 1919 | td_zone_free_index(td); |
7c4d0fb7 | 1920 | |
b2a9e649 | 1921 | if (fio_option_is_set(o, cpumask)) { |
8a1db9a1 JA |
1922 | ret = fio_cpuset_exit(&o->cpumask); |
1923 | if (ret) | |
1924 | td_verror(td, ret, "fio_cpuset_exit"); | |
2e1df07d JA |
1925 | } |
1926 | ||
1927 | /* | |
1928 | * do this very late, it will log file closing as well | |
1929 | */ | |
4896473e | 1930 | if (o->write_iolog_file) |
2e1df07d | 1931 | write_iolog_close(td); |
98e7161c AK |
1932 | if (td->io_log_rfile) |
1933 | fclose(td->io_log_rfile); | |
2e1df07d JA |
1934 | |
1935 | td_set_runstate(td, TD_EXITED); | |
fda2cfac JA |
1936 | |
1937 | /* | |
1938 | * Do this last after setting our runstate to exited, so we | |
1939 | * know that the stat thread is signaled. | |
1940 | */ | |
1941 | check_update_rusage(td); | |
1942 | ||
24660963 | 1943 | sk_out_drop(); |
e43606c2 | 1944 | return (void *) (uintptr_t) td->error; |
2e1df07d JA |
1945 | } |
1946 | ||
2e1df07d JA |
1947 | /* |
1948 | * Run over the job map and reap the threads that have exited, if any. | |
1949 | */ | |
90eff1c9 SW |
1950 | static void reap_threads(unsigned int *nr_running, uint64_t *t_rate, |
1951 | uint64_t *m_rate) | |
2e1df07d JA |
1952 | { |
1953 | struct thread_data *td; | |
1954 | unsigned int cputhreads, realthreads, pending; | |
1955 | int i, status, ret; | |
1956 | ||
1957 | /* | |
1958 | * reap exited threads (TD_EXITED -> TD_REAPED) | |
1959 | */ | |
1960 | realthreads = pending = cputhreads = 0; | |
1961 | for_each_td(td, i) { | |
1962 | int flags = 0; | |
1963 | ||
8847ae4c | 1964 | if (!strcmp(td->o.ioengine, "cpuio")) |
2e1df07d JA |
1965 | cputhreads++; |
1966 | else | |
1967 | realthreads++; | |
1968 | ||
1969 | if (!td->pid) { | |
1970 | pending++; | |
1971 | continue; | |
1972 | } | |
1973 | if (td->runstate == TD_REAPED) | |
1974 | continue; | |
1975 | if (td->o.use_thread) { | |
1976 | if (td->runstate == TD_EXITED) { | |
1977 | td_set_runstate(td, TD_REAPED); | |
1978 | goto reaped; | |
1979 | } | |
1980 | continue; | |
1981 | } | |
1982 | ||
1983 | flags = WNOHANG; | |
1984 | if (td->runstate == TD_EXITED) | |
1985 | flags = 0; | |
1986 | ||
1987 | /* | |
1988 | * check if someone quit or got killed in an unusual way | |
1989 | */ | |
1990 | ret = waitpid(td->pid, &status, flags); | |
1991 | if (ret < 0) { | |
1992 | if (errno == ECHILD) { | |
1993 | log_err("fio: pid=%d disappeared %d\n", | |
1994 | (int) td->pid, td->runstate); | |
a5e371a6 | 1995 | td->sig = ECHILD; |
2e1df07d JA |
1996 | td_set_runstate(td, TD_REAPED); |
1997 | goto reaped; | |
1998 | } | |
1999 | perror("waitpid"); | |
2000 | } else if (ret == td->pid) { | |
2001 | if (WIFSIGNALED(status)) { | |
2002 | int sig = WTERMSIG(status); | |
2003 | ||
36d80bc7 | 2004 | if (sig != SIGTERM && sig != SIGUSR2) |
2e1df07d JA |
2005 | log_err("fio: pid=%d, got signal=%d\n", |
2006 | (int) td->pid, sig); | |
a5e371a6 | 2007 | td->sig = sig; |
2e1df07d JA |
2008 | td_set_runstate(td, TD_REAPED); |
2009 | goto reaped; | |
2010 | } | |
2011 | if (WIFEXITED(status)) { | |
2012 | if (WEXITSTATUS(status) && !td->error) | |
2013 | td->error = WEXITSTATUS(status); | |
2014 | ||
2015 | td_set_runstate(td, TD_REAPED); | |
2016 | goto reaped; | |
2017 | } | |
2018 | } | |
2019 | ||
cba5460c JA |
2020 | /* |
2021 | * If the job is stuck, do a forceful timeout of it and | |
2022 | * move on. | |
2023 | */ | |
2024 | if (td->terminate && | |
2c45a4ac | 2025 | td->runstate < TD_FSYNCING && |
cba5460c | 2026 | time_since_now(&td->terminate_time) >= FIO_REAP_TIMEOUT) { |
86310a1f TK |
2027 | log_err("fio: job '%s' (state=%d) hasn't exited in " |
2028 | "%lu seconds, it appears to be stuck. Doing " | |
2029 | "forceful exit of this job.\n", | |
2030 | td->o.name, td->runstate, | |
2031 | (unsigned long) time_since_now(&td->terminate_time)); | |
cba5460c JA |
2032 | td_set_runstate(td, TD_REAPED); |
2033 | goto reaped; | |
2034 | } | |
2035 | ||
2e1df07d JA |
2036 | /* |
2037 | * thread is not dead, continue | |
2038 | */ | |
2039 | pending++; | |
2040 | continue; | |
2041 | reaped: | |
2042 | (*nr_running)--; | |
342f4be4 JA |
2043 | (*m_rate) -= ddir_rw_sum(td->o.ratemin); |
2044 | (*t_rate) -= ddir_rw_sum(td->o.rate); | |
2e1df07d JA |
2045 | if (!td->pid) |
2046 | pending--; | |
2047 | ||
2048 | if (td->error) | |
2049 | exit_value++; | |
2050 | ||
2051 | done_secs += mtime_since_now(&td->epoch) / 1000; | |
4a88752a | 2052 | profile_td_exit(td); |
42e26e94 | 2053 | flow_exit_job(td); |
2e1df07d JA |
2054 | } |
2055 | ||
2056 | if (*nr_running == cputhreads && !pending && realthreads) | |
64402a8a | 2057 | fio_terminate_threads(TERMINATE_ALL, TERMINATE_ALL); |
2e1df07d JA |
2058 | } |
2059 | ||
e39c0676 | 2060 | static bool __check_trigger_file(void) |
ca09be4b JA |
2061 | { |
2062 | struct stat sb; | |
2063 | ||
2064 | if (!trigger_file) | |
e39c0676 | 2065 | return false; |
ca09be4b JA |
2066 | |
2067 | if (stat(trigger_file, &sb)) | |
e39c0676 | 2068 | return false; |
ca09be4b JA |
2069 | |
2070 | if (unlink(trigger_file) < 0) | |
2071 | log_err("fio: failed to unlink %s: %s\n", trigger_file, | |
2072 | strerror(errno)); | |
2073 | ||
e39c0676 | 2074 | return true; |
ca09be4b JA |
2075 | } |
2076 | ||
e39c0676 | 2077 | static bool trigger_timedout(void) |
ca09be4b JA |
2078 | { |
2079 | if (trigger_timeout) | |
98413be7 SW |
2080 | if (time_since_genesis() >= trigger_timeout) { |
2081 | trigger_timeout = 0; | |
2082 | return true; | |
2083 | } | |
ca09be4b | 2084 | |
e39c0676 | 2085 | return false; |
ca09be4b JA |
2086 | } |
2087 | ||
2088 | void exec_trigger(const char *cmd) | |
2089 | { | |
2090 | int ret; | |
2091 | ||
98413be7 | 2092 | if (!cmd || cmd[0] == '\0') |
ca09be4b JA |
2093 | return; |
2094 | ||
2095 | ret = system(cmd); | |
2096 | if (ret == -1) | |
2097 | log_err("fio: failed executing %s trigger\n", cmd); | |
2098 | } | |
2099 | ||
2100 | void check_trigger_file(void) | |
2101 | { | |
2102 | if (__check_trigger_file() || trigger_timedout()) { | |
796fb3ce JA |
2103 | if (nr_clients) |
2104 | fio_clients_send_trigger(trigger_remote_cmd); | |
2105 | else { | |
d264264a | 2106 | verify_save_state(IO_LIST_ALL); |
64402a8a | 2107 | fio_terminate_threads(TERMINATE_ALL, TERMINATE_ALL); |
ca09be4b JA |
2108 | exec_trigger(trigger_cmd); |
2109 | } | |
2110 | } | |
2111 | } | |
2112 | ||
2113 | static int fio_verify_load_state(struct thread_data *td) | |
2114 | { | |
2115 | int ret; | |
2116 | ||
2117 | if (!td->o.verify_state) | |
2118 | return 0; | |
2119 | ||
2120 | if (is_backend) { | |
2121 | void *data; | |
2122 | ||
2123 | ret = fio_server_get_verify_state(td->o.name, | |
94a6e1bb | 2124 | td->thread_number - 1, &data); |
ca09be4b | 2125 | if (!ret) |
94a6e1bb | 2126 | verify_assign_state(td, data); |
67710881 AK |
2127 | } else { |
2128 | char prefix[PATH_MAX]; | |
2129 | ||
2130 | if (aux_path) | |
2131 | sprintf(prefix, "%s%clocal", aux_path, | |
2132 | FIO_OS_PATH_SEPARATOR); | |
2133 | else | |
2134 | strcpy(prefix, "local"); | |
2135 | ret = verify_load_state(td, prefix); | |
2136 | } | |
ca09be4b JA |
2137 | |
2138 | return ret; | |
2139 | } | |
2140 | ||
06464907 JA |
2141 | static void do_usleep(unsigned int usecs) |
2142 | { | |
2143 | check_for_running_stats(); | |
ca09be4b | 2144 | check_trigger_file(); |
06464907 JA |
2145 | usleep(usecs); |
2146 | } | |
2147 | ||
e39c0676 | 2148 | static bool check_mount_writes(struct thread_data *td) |
e81ecca3 JA |
2149 | { |
2150 | struct fio_file *f; | |
2151 | unsigned int i; | |
2152 | ||
2153 | if (!td_write(td) || td->o.allow_mounted_write) | |
e39c0676 | 2154 | return false; |
e81ecca3 | 2155 | |
5e81f9c1 TK |
2156 | /* |
2157 | * If FIO_HAVE_CHARDEV_SIZE is defined, it's likely that chrdevs | |
2158 | * are mkfs'd and mounted. | |
2159 | */ | |
e81ecca3 | 2160 | for_each_file(td, f, i) { |
5e81f9c1 TK |
2161 | #ifdef FIO_HAVE_CHARDEV_SIZE |
2162 | if (f->filetype != FIO_TYPE_BLOCK && f->filetype != FIO_TYPE_CHAR) | |
2163 | #else | |
686fbd31 | 2164 | if (f->filetype != FIO_TYPE_BLOCK) |
5e81f9c1 | 2165 | #endif |
e81ecca3 JA |
2166 | continue; |
2167 | if (device_is_mounted(f->file_name)) | |
2168 | goto mounted; | |
2169 | } | |
2170 | ||
e39c0676 | 2171 | return false; |
e81ecca3 | 2172 | mounted: |
c5839011 | 2173 | log_err("fio: %s appears mounted, and 'allow_mounted_write' isn't set. Aborting.\n", f->file_name); |
e39c0676 | 2174 | return true; |
e81ecca3 JA |
2175 | } |
2176 | ||
9cc8cb91 AK |
2177 | static bool waitee_running(struct thread_data *me) |
2178 | { | |
2179 | const char *waitee = me->o.wait_for; | |
2180 | const char *self = me->o.name; | |
2181 | struct thread_data *td; | |
2182 | int i; | |
2183 | ||
2184 | if (!waitee) | |
2185 | return false; | |
2186 | ||
2187 | for_each_td(td, i) { | |
2188 | if (!strcmp(td->o.name, self) || strcmp(td->o.name, waitee)) | |
2189 | continue; | |
2190 | ||
2191 | if (td->runstate < TD_EXITED) { | |
2192 | dprint(FD_PROCESS, "%s fenced by %s(%s)\n", | |
2193 | self, td->o.name, | |
2194 | runstate_to_name(td->runstate)); | |
2195 | return true; | |
2196 | } | |
2197 | } | |
2198 | ||
2199 | dprint(FD_PROCESS, "%s: %s completed, can run\n", self, waitee); | |
2200 | return false; | |
2201 | } | |
2202 | ||
2e1df07d JA |
2203 | /* |
2204 | * Main function for kicking off and reaping jobs, as needed. | |
2205 | */ | |
24660963 | 2206 | static void run_threads(struct sk_out *sk_out) |
2e1df07d JA |
2207 | { |
2208 | struct thread_data *td; | |
90eff1c9 SW |
2209 | unsigned int i, todo, nr_running, nr_started; |
2210 | uint64_t m_rate, t_rate; | |
0de5b26f | 2211 | uint64_t spent; |
2e1df07d | 2212 | |
2e1df07d JA |
2213 | if (fio_gtod_offload && fio_start_gtod_thread()) |
2214 | return; | |
334185e9 | 2215 | |
f2a2ce0e | 2216 | fio_idle_prof_init(); |
2e1df07d JA |
2217 | |
2218 | set_sig_handlers(); | |
2219 | ||
3a5f6bde JA |
2220 | nr_thread = nr_process = 0; |
2221 | for_each_td(td, i) { | |
e81ecca3 JA |
2222 | if (check_mount_writes(td)) |
2223 | return; | |
3a5f6bde JA |
2224 | if (td->o.use_thread) |
2225 | nr_thread++; | |
2226 | else | |
2227 | nr_process++; | |
2228 | } | |
2229 | ||
01cfefcc | 2230 | if (output_format & FIO_OUTPUT_NORMAL) { |
a41fc529 JA |
2231 | struct buf_output out; |
2232 | ||
2233 | buf_output_init(&out); | |
c2fcb6bc | 2234 | __log_buf(&out, "Starting "); |
2e1df07d | 2235 | if (nr_thread) |
c2fcb6bc | 2236 | __log_buf(&out, "%d thread%s", nr_thread, |
2e1df07d JA |
2237 | nr_thread > 1 ? "s" : ""); |
2238 | if (nr_process) { | |
2239 | if (nr_thread) | |
c2fcb6bc JA |
2240 | __log_buf(&out, " and "); |
2241 | __log_buf(&out, "%d process%s", nr_process, | |
2e1df07d JA |
2242 | nr_process > 1 ? "es" : ""); |
2243 | } | |
c2fcb6bc | 2244 | __log_buf(&out, "\n"); |
a41fc529 | 2245 | log_info_buf(out.buf, out.buflen); |
d5bb0a89 | 2246 | buf_output_free(&out); |
2e1df07d JA |
2247 | } |
2248 | ||
2249 | todo = thread_number; | |
2250 | nr_running = 0; | |
2251 | nr_started = 0; | |
2252 | m_rate = t_rate = 0; | |
2253 | ||
2254 | for_each_td(td, i) { | |
2255 | print_status_init(td->thread_number - 1); | |
2256 | ||
2257 | if (!td->o.create_serialize) | |
2258 | continue; | |
2259 | ||
ca09be4b JA |
2260 | if (fio_verify_load_state(td)) |
2261 | goto reap; | |
2262 | ||
2e1df07d JA |
2263 | /* |
2264 | * do file setup here so it happens sequentially, | |
2265 | * we don't want X number of threads getting their | |
2266 | * client data interspersed on disk | |
2267 | */ | |
2268 | if (setup_files(td)) { | |
ca09be4b | 2269 | reap: |
2e1df07d JA |
2270 | exit_value++; |
2271 | if (td->error) | |
2272 | log_err("fio: pid=%d, err=%d/%s\n", | |
2273 | (int) td->pid, td->error, td->verror); | |
2274 | td_set_runstate(td, TD_REAPED); | |
2275 | todo--; | |
2276 | } else { | |
2277 | struct fio_file *f; | |
2278 | unsigned int j; | |
2279 | ||
2280 | /* | |
2281 | * for sharing to work, each job must always open | |
2282 | * its own files. so close them, if we opened them | |
2283 | * for creation | |
2284 | */ | |
2285 | for_each_file(td, f, j) { | |
2286 | if (fio_file_open(f)) | |
2287 | td_io_close_file(td, f); | |
2288 | } | |
2289 | } | |
2290 | } | |
2291 | ||
f2a2ce0e HL |
2292 | /* start idle threads before io threads start to run */ |
2293 | fio_idle_prof_start(); | |
2294 | ||
2e1df07d JA |
2295 | set_genesis_time(); |
2296 | ||
2297 | while (todo) { | |
2298 | struct thread_data *map[REAL_MAX_JOBS]; | |
8b6a404c | 2299 | struct timespec this_start; |
2e1df07d | 2300 | int this_jobs = 0, left; |
ce8b6139 | 2301 | struct fork_data *fd; |
2e1df07d JA |
2302 | |
2303 | /* | |
2304 | * create threads (TD_NOT_CREATED -> TD_CREATED) | |
2305 | */ | |
2306 | for_each_td(td, i) { | |
2307 | if (td->runstate != TD_NOT_CREATED) | |
2308 | continue; | |
2309 | ||
2310 | /* | |
2311 | * never got a chance to start, killed by other | |
2312 | * thread for some reason | |
2313 | */ | |
2314 | if (td->terminate) { | |
2315 | todo--; | |
2316 | continue; | |
2317 | } | |
2318 | ||
2319 | if (td->o.start_delay) { | |
0de5b26f | 2320 | spent = utime_since_genesis(); |
2e1df07d | 2321 | |
74454ce4 | 2322 | if (td->o.start_delay > spent) |
2e1df07d JA |
2323 | continue; |
2324 | } | |
2325 | ||
2326 | if (td->o.stonewall && (nr_started || nr_running)) { | |
2327 | dprint(FD_PROCESS, "%s: stonewall wait\n", | |
2328 | td->o.name); | |
2329 | break; | |
2330 | } | |
2331 | ||
9cc8cb91 AK |
2332 | if (waitee_running(td)) { |
2333 | dprint(FD_PROCESS, "%s: waiting for %s\n", | |
2334 | td->o.name, td->o.wait_for); | |
5c74fc76 | 2335 | continue; |
9cc8cb91 AK |
2336 | } |
2337 | ||
2e1df07d JA |
2338 | init_disk_util(td); |
2339 | ||
971caeb1 | 2340 | td->rusage_sem = fio_sem_init(FIO_SEM_LOCKED); |
c97f1ad6 JA |
2341 | td->update_rusage = 0; |
2342 | ||
2e1df07d JA |
2343 | /* |
2344 | * Set state to created. Thread will transition | |
2345 | * to TD_INITIALIZED when it's done setting up. | |
2346 | */ | |
2347 | td_set_runstate(td, TD_CREATED); | |
2348 | map[this_jobs++] = td; | |
2349 | nr_started++; | |
2350 | ||
ce8b6139 JA |
2351 | fd = calloc(1, sizeof(*fd)); |
2352 | fd->td = td; | |
2353 | fd->sk_out = sk_out; | |
2354 | ||
2e1df07d JA |
2355 | if (td->o.use_thread) { |
2356 | int ret; | |
2357 | ||
2358 | dprint(FD_PROCESS, "will pthread_create\n"); | |
2359 | ret = pthread_create(&td->thread, NULL, | |
24660963 | 2360 | thread_main, fd); |
2e1df07d JA |
2361 | if (ret) { |
2362 | log_err("pthread_create: %s\n", | |
2363 | strerror(ret)); | |
24660963 | 2364 | free(fd); |
2e1df07d JA |
2365 | nr_started--; |
2366 | break; | |
2367 | } | |
493dd6c1 | 2368 | fd = NULL; |
2e1df07d JA |
2369 | ret = pthread_detach(td->thread); |
2370 | if (ret) | |
2371 | log_err("pthread_detach: %s", | |
2372 | strerror(ret)); | |
2373 | } else { | |
2374 | pid_t pid; | |
2375 | dprint(FD_PROCESS, "will fork\n"); | |
2376 | pid = fork(); | |
2377 | if (!pid) { | |
ce8b6139 | 2378 | int ret; |
2e1df07d | 2379 | |
ce8b6139 | 2380 | ret = (int)(uintptr_t)thread_main(fd); |
2e1df07d JA |
2381 | _exit(ret); |
2382 | } else if (i == fio_debug_jobno) | |
2383 | *fio_debug_jobp = pid; | |
2384 | } | |
971caeb1 BVA |
2385 | dprint(FD_MUTEX, "wait on startup_sem\n"); |
2386 | if (fio_sem_down_timeout(startup_sem, 10000)) { | |
2e1df07d | 2387 | log_err("fio: job startup hung? exiting.\n"); |
64402a8a | 2388 | fio_terminate_threads(TERMINATE_ALL, TERMINATE_ALL); |
f1867a7f | 2389 | fio_abort = true; |
2e1df07d | 2390 | nr_started--; |
6541f4e0 | 2391 | free(fd); |
2e1df07d JA |
2392 | break; |
2393 | } | |
971caeb1 | 2394 | dprint(FD_MUTEX, "done waiting on startup_sem\n"); |
2e1df07d JA |
2395 | } |
2396 | ||
2397 | /* | |
2398 | * Wait for the started threads to transition to | |
2399 | * TD_INITIALIZED. | |
2400 | */ | |
2401 | fio_gettime(&this_start, NULL); | |
2402 | left = this_jobs; | |
2403 | while (left && !fio_abort) { | |
2404 | if (mtime_since_now(&this_start) > JOB_START_TIMEOUT) | |
2405 | break; | |
2406 | ||
06464907 | 2407 | do_usleep(100000); |
2e1df07d JA |
2408 | |
2409 | for (i = 0; i < this_jobs; i++) { | |
2410 | td = map[i]; | |
2411 | if (!td) | |
2412 | continue; | |
2413 | if (td->runstate == TD_INITIALIZED) { | |
2414 | map[i] = NULL; | |
2415 | left--; | |
2416 | } else if (td->runstate >= TD_EXITED) { | |
2417 | map[i] = NULL; | |
2418 | left--; | |
2419 | todo--; | |
2420 | nr_running++; /* work-around... */ | |
2421 | } | |
2422 | } | |
2423 | } | |
2424 | ||
2425 | if (left) { | |
4e87c37a JA |
2426 | log_err("fio: %d job%s failed to start\n", left, |
2427 | left > 1 ? "s" : ""); | |
2e1df07d JA |
2428 | for (i = 0; i < this_jobs; i++) { |
2429 | td = map[i]; | |
2430 | if (!td) | |
2431 | continue; | |
2432 | kill(td->pid, SIGTERM); | |
2433 | } | |
2434 | break; | |
2435 | } | |
2436 | ||
2437 | /* | |
2438 | * start created threads (TD_INITIALIZED -> TD_RUNNING). | |
2439 | */ | |
2440 | for_each_td(td, i) { | |
2441 | if (td->runstate != TD_INITIALIZED) | |
2442 | continue; | |
2443 | ||
2444 | if (in_ramp_time(td)) | |
2445 | td_set_runstate(td, TD_RAMP); | |
2446 | else | |
2447 | td_set_runstate(td, TD_RUNNING); | |
2448 | nr_running++; | |
2449 | nr_started--; | |
342f4be4 JA |
2450 | m_rate += ddir_rw_sum(td->o.ratemin); |
2451 | t_rate += ddir_rw_sum(td->o.rate); | |
2e1df07d | 2452 | todo--; |
971caeb1 | 2453 | fio_sem_up(td->sem); |
2e1df07d JA |
2454 | } |
2455 | ||
2456 | reap_threads(&nr_running, &t_rate, &m_rate); | |
2457 | ||
122c7725 | 2458 | if (todo) |
06464907 | 2459 | do_usleep(100000); |
2e1df07d JA |
2460 | } |
2461 | ||
2462 | while (nr_running) { | |
2463 | reap_threads(&nr_running, &t_rate, &m_rate); | |
06464907 | 2464 | do_usleep(10000); |
2e1df07d JA |
2465 | } |
2466 | ||
f2a2ce0e HL |
2467 | fio_idle_prof_stop(); |
2468 | ||
2e1df07d | 2469 | update_io_ticks(); |
2e1df07d JA |
2470 | } |
2471 | ||
27357187 JA |
2472 | static void free_disk_util(void) |
2473 | { | |
27357187 | 2474 | disk_util_prune_entries(); |
a39fb9ea | 2475 | helper_thread_destroy(); |
2e1df07d JA |
2476 | } |
2477 | ||
24660963 | 2478 | int fio_backend(struct sk_out *sk_out) |
2e1df07d JA |
2479 | { |
2480 | struct thread_data *td; | |
2481 | int i; | |
2482 | ||
2483 | if (exec_profile) { | |
2484 | if (load_profile(exec_profile)) | |
2485 | return 1; | |
2486 | free(exec_profile); | |
2487 | exec_profile = NULL; | |
2488 | } | |
2489 | if (!thread_number) | |
2490 | return 0; | |
2491 | ||
2492 | if (write_bw_log) { | |
aee2ab67 JA |
2493 | struct log_params p = { |
2494 | .log_type = IO_LOG_TYPE_BW, | |
2495 | }; | |
2496 | ||
2497 | setup_log(&agg_io_log[DDIR_READ], &p, "agg-read_bw.log"); | |
2498 | setup_log(&agg_io_log[DDIR_WRITE], &p, "agg-write_bw.log"); | |
2499 | setup_log(&agg_io_log[DDIR_TRIM], &p, "agg-trim_bw.log"); | |
2e1df07d JA |
2500 | } |
2501 | ||
971caeb1 | 2502 | startup_sem = fio_sem_init(FIO_SEM_LOCKED); |
24a97130 LC |
2503 | if (!sk_out) |
2504 | is_local_backend = true; | |
971caeb1 | 2505 | if (startup_sem == NULL) |
2e1df07d | 2506 | return 1; |
2e1df07d JA |
2507 | |
2508 | set_genesis_time(); | |
cef9175e | 2509 | stat_init(); |
998e9ebb BVA |
2510 | if (helper_thread_create(startup_sem, sk_out)) |
2511 | log_err("fio: failed to create helper thread\n"); | |
2e1df07d JA |
2512 | |
2513 | cgroup_list = smalloc(sizeof(*cgroup_list)); | |
b2e6494c JA |
2514 | if (cgroup_list) |
2515 | INIT_FLIST_HEAD(cgroup_list); | |
2e1df07d | 2516 | |
24660963 | 2517 | run_threads(sk_out); |
2e1df07d | 2518 | |
a39fb9ea | 2519 | helper_thread_exit(); |
8aab824f | 2520 | |
2e1df07d | 2521 | if (!fio_abort) { |
83f7b64e | 2522 | __show_run_stats(); |
2e1df07d | 2523 | if (write_bw_log) { |
cb7e0ace JA |
2524 | for (i = 0; i < DDIR_RWDIR_CNT; i++) { |
2525 | struct io_log *log = agg_io_log[i]; | |
2526 | ||
60a25727 | 2527 | flush_log(log, false); |
518dac09 | 2528 | free_log(log); |
cb7e0ace | 2529 | } |
2e1df07d JA |
2530 | } |
2531 | } | |
2532 | ||
fda2cfac | 2533 | for_each_td(td, i) { |
c27cc65f | 2534 | steadystate_free(td); |
2e1df07d | 2535 | fio_options_free(td); |
8049adc1 | 2536 | if (td->rusage_sem) { |
971caeb1 | 2537 | fio_sem_remove(td->rusage_sem); |
8049adc1 JA |
2538 | td->rusage_sem = NULL; |
2539 | } | |
971caeb1 BVA |
2540 | fio_sem_remove(td->sem); |
2541 | td->sem = NULL; | |
fda2cfac | 2542 | } |
2e1df07d | 2543 | |
a462baef | 2544 | free_disk_util(); |
b2e6494c JA |
2545 | if (cgroup_list) { |
2546 | cgroup_kill(cgroup_list); | |
2547 | sfree(cgroup_list); | |
2548 | } | |
2e1df07d | 2549 | |
971caeb1 | 2550 | fio_sem_remove(startup_sem); |
cef9175e | 2551 | stat_exit(); |
2e1df07d JA |
2552 | return exit_value; |
2553 | } |